Why do changes in the same gene sometimes raise BMI and sometimes lower it? The clearest place to start is appetite biology: MC4R links variant function to central energy-balance signaling, while PCSK1 connects human obesity genetics to hormone processing. Our completed analysis separates these biological connections from three distinct statistical questions, using 7.96 million variant–gene records rather than treating every allele as equivalent.
An annotation meta-regression of 7,959,900 variant–gene rows across 18,752 retrieved genes asks whether predicted molecular consequences explain differences in BMI effects. Five genes cross the prespecified annotation-family threshold: MC4R, BLTP1/KIAA1109, UBR2, ADARB1 and DIDO1. The variant-level evidence is informative, but its interpretation depends on which statistical question is being asked and how much information individual alleles contribute.
The fixed universe contains 19,287 genes from the BMI pLoF, missense|LC and synonymous source tables. Collection is finalized under the documented stopping rule, including one supplementary attempt for each initially unavailable source. There are 535 unavailable effect sources and 2 unavailable QC sources, with zero retries pending. Finalized available-source analysis does not mean complete source coverage. Missing effects were never converted into empty genes or imputed betas.
Sensitivity checks materially qualify two of the five annotation results: DIDO1 loses its planned-family significance after certain single-variant deletions, and ADARB1 weakens substantially when a local cluster of identical-summary records is removed. These descriptive checks leave the prespecified primary model unchanged and motivate genotype-level validation.
This is a secondary analysis of the public Genebass summary statistics for continuous-BMI_custom-both_sexes--custom, with 393,562 analyzed participants. Source metadata records inv_normalized=False. Effect sizes below are on the Genebass BMI analysis scale; we do not relabel them as kilograms, per-carrier effects or clinical treatment effects.
Appetite and feeding: the genes to read first
The strongest biological connections span central appetite signaling, prohormone processing and incretin signaling. Their statistical evidence is different, so the table keeps the primary annotation family separate from the added overall and centered-mean tests.
| Gene | Connection to appetite or feeding | What passes in this analysis? | Key qualification |
|---|---|---|---|
| MC4R | Central melanocortin signaling; human loss- and gain-of-function alleles have contrasting BMI associations. Human functional/genetic evidence. | Primary annotation p=3.56×10−16; all three baseline tests also pass the conservative correction. | Protective and BMI-increasing alleles coexist. No single-variant deletion removes the planned annotation significance. |
| PCSK1 | Prohormone convertase 1/3 links hormone processing to established human obesity genetics. Human exome evidence. | Overall p=8.34×10−12; centered p=4.65×10−10, both conservatively significant. | Annotation p=1.47×10−4 does not pass the planned annotation threshold. One influential-variant deletion removes the centered test’s per-test significance. |
| ADARB1 / ADAR2 | Misexpression causes hyperphagia and obesity in transgenic mice. Experimental feeding evidence. | Primary annotation p=3.50×10−7. | The mouse experiment does not establish the human mechanism. Removing nine adjacent identical-summary records changes annotation p to 0.0455. |
| GIPR | Incretin receptor connecting meal-related hormone biology to human coding-variant BMI evidence. Functional/genetic evidence. | Centered p=2.08×10−6 passes the 19,287-gene per-test correction. | It does not pass the conservative three-family correction (adjusted p≈0.12), and neither overall nor annotation testing is a planned-family hit. |
The distinction matters: these BMI summary statistics do not measure food intake, hunger or satiety directly. Established appetite biology helps interpret the associations; the present analysis does not prove that a given human allele changes feeding behavior, identify a causal tissue, or predict the effect of a drug. No pathway or cell-type enrichment survives the prespecified multiple-testing corrections.
Protective variants: genes with evidence of lower BMI
Some alleles point toward lower BMI rather than higher BMI. MC4R offers the clearest examples with published human and functional support; GIPR supplies additional BMI-lowering missense alleles. GPR151 illustrates why an observed protective association is not enough to establish a drug target. Here, protective refers to the direction of the BMI association, not proven causality or a guarantee of better overall health.
MC4R: protective alleles within a gene that also contains obesity-risk variants
Val103Ile (V103I) and Ile251Leu (I251L) both have negative BMI estimates in the frozen Genebass data. Published experiments describe gain of function and altered receptor signaling/trafficking, including signaling biased toward β-arrestin recruitment. These alleles provide a useful contrast to MC4R loss-of-function variants associated with higher BMI. The functional interpretation is assay-specific; a single annotation score cannot represent every receptor effect. Lotta et al., Cell 2019; MC4R trafficking experiments.
V103I has particularly strong single-variant evidence here (β=−0.1070, source p=1.71×10−43). I251L has a smaller estimate (β=−0.0267, source p=0.00394); its current single-variant evidence is nominal despite the supporting prior literature. Removing V103I changes MC4R’s centered-mean p from 3.76×10−10 to 0.798, while its original annotation F remains strongly significant (p=2.94×10−14). Thus the downward weighted mean and the annotation-related differences have distinct evidential support. See the MC4R variant panel and full sensitivity analysis.
GIPR: BMI-lowering missense variants with functional evidence
Glu288Gly (E288G) and Arg190Gln (R190Q) have negative BMI estimates and published evidence for impaired receptor signaling, including effects on cAMP and β-arrestin pathways. The molecular loss of receptor function described in those experiments is distinct from the binary HC-LoF annotation used in this regression: both are missense variants with observed non-HC labels here. Variant pharmacology and human associations; Kizilkaya et al., Nature Metabolism 2024.
GIPR passes the planned centered-mean family (p=2.08×10−6) but not the conservative three-family correction (adjusted p=0.120). Removing E288G or R190Q weakens centered-mean p to 0.000349 or 0.0000556, respectively. The annotation F is not significant (p=0.468). These results support examining particular BMI-lowering alleles; they do not establish that every disruptive GIPR allele has the same effect or determine the clinical effects of agonism versus antagonism. See the GIPR variant panel.
GPR151: a protective association with an important qualification
The truncating Arg95Ter (R95*) allele has β=−0.0628 and source p=1.41×10−7. GPR151’s centered-mean p=1.51×10−6 passes the per-test planned correction, but not the conservative three-family correction (adjusted p=0.0875); removing Arg95Ter changes that mean-test p to 0.0556. AlphaMissense is missing and imputed for this allele, while its HC-LoF label is observed.
Prior human studies reported a BMI-lowering Arg95Ter association, but follow-up studies of complete human knockouts did not establish a clinically meaningful BMI reduction. This is a useful example of the gap between an allele association and a therapeutic conclusion. Initial protective-variant study; Human knockout follow-up. See the GPR151 variant panel.
The BMI-lowering alleles in the frozen inputs
| Gene | Allele · GRCh38 variant | Variant β · approximate 95% interval | Source variant p | Allele frequency |
|---|---|---|---|---|
| MC4R | V103I18-60372043-C-T | -0.107 [-0.1222, -0.09187] | 1.71e-43 | 2.01% |
| MC4R | I251L18-60371599-T-G | -0.02666 [-0.04479, -0.008538] | 0.00394 | 1.4% |
| GIPR | E288G19-45677718-A-G | -0.1119 [-0.1608, -0.06294] | 7.41e-06 | 0.189% |
| GIPR | R190Q19-45674762-G-A | -0.08689 [-0.1413, -0.03249] | 0.00174 | 0.153% |
| GPR151 | R95*5-146515831-G-A | -0.0628 [-0.08618, -0.03942] | 1.41e-07 | 0.831% |
These five alleles were selected for illustration using the frozen inputs and published biology. Variant p-values are unadjusted source p-values, separate from the corrected gene-level tests. The displayed variant intervals are β±1.96×SE, using the same native-or-reconstructed SE as the original analysis; they are not the residual-scaled intervals for gene-level moderator contrasts. Frequencies are allele frequencies, not carrier percentages. Published association samples may overlap Genebass/UK Biobank, so this comparison is not automatically independent replication.
Broader downward signals: 19 genes, with different levels of interpretation
Of the 40 planned-family centered-mean hits, 19 have negative estimates; 15 of those also pass the conservative 57,861-hypothesis correction. These estimates summarize the inverse-variance-weighted baseline variant effects. They are not gene-knockout, LoF burden or per-carrier effects. The original annotation F and overall F have no signed direction by themselves.
View all 19 genes with significantly negative centered means
| Gene | Centered mean · 95% t interval | Centered p | 19,287-gene adjusted p | Also passes 57,861? |
|---|---|---|---|---|
| TNKS1BP1 | -0.0156 [-0.019, -0.01219] | 7.63e-19 | 1.47e-14 | Yes |
| ANKRD11 | -0.01256 [-0.0163, -0.008815] | 5.46e-11 | 1.05e-06 | Yes |
| CDH15 | -0.02047 [-0.02667, -0.01427] | 1.46e-10 | 2.82e-06 | Yes |
| MC4R | -0.05091 [-0.06628, -0.03555] | 3.76e-10 | 7.25e-06 | Yes |
| LRP3 | -0.01437 [-0.01908, -0.009647] | 3.4e-09 | 6.56e-05 | Yes |
| PABPC4 | -0.03501 [-0.04653, -0.02349] | 4.72e-09 | 9.11e-05 | Yes |
| CPNE1 | -0.01943 [-0.02606, -0.0128] | 1.62e-08 | 0.000312 | Yes |
| PRRC2A | -0.01154 [-0.01553, -0.007544] | 1.64e-08 | 0.000317 | Yes |
| NUTM2F | -0.01565 [-0.02115, -0.01014] | 3.75e-08 | 0.000723 | Yes |
| CCDC171 | -0.01538 [-0.02108, -0.00969] | 1.36e-07 | 0.00263 | Yes |
| MTARC2 | -0.04368 [-0.05985, -0.02751] | 2.35e-07 | 0.00453 | Yes |
| VPS13C | -0.01255 [-0.01732, -0.007782] | 2.67e-07 | 0.00515 | Yes |
| SEMA3G | -0.02173 [-0.03015, -0.01331] | 5.11e-07 | 0.00985 | Yes |
| TRPM4 | -0.01416 [-0.01973, -0.008581] | 7.31e-07 | 0.0141 | Yes |
| SLC22A31 | -0.01964 [-0.02733, -0.01195] | 7.91e-07 | 0.0153 | Yes |
| NUP214 | -0.01664 [-0.02334, -0.009933] | 1.25e-06 | 0.024 | Per-test only |
| GPR151 | -0.04417 [-0.06192, -0.02642] | 1.51e-06 | 0.0292 | Per-test only |
| GIPR | -0.03929 [-0.05532, -0.02325] | 2.08e-06 | 0.0401 | Per-test only |
| MST1R | -0.02245 [-0.03173, -0.01317] | 2.32e-06 | 0.0448 | Per-test only |
For the less established genes, a downward signal may reflect a tagging allele rather than a protective protein mechanism. For example, the synonymous PABPC4 Ile155= allele has β=−0.0383; removing it changes centered-mean p from 4.72×10−9 to 0.982. Such sensitivity motivates allele-level and genotype-level follow-up rather than a claim that inhibiting the gene will lower BMI. All original variants, imputation values, QC flags and fitted tests remain unchanged.
Three questions, three tests
| Baseline joint-model test | Testable genes | Bonferroni:19,287 per test | Conservative:57,861 across tests |
|---|---|---|---|
| Annotation F | 18,703 | 5 | 5 |
| Overall F | 18,736 | 32 | 25 |
| Centered-mean t | 18,736 | 40 | 31 |
The original annotation F test asks whether adding HMM, AlphaMissense and HC-LoF improves the fit beyond an intercept. The overall F test asks whether the full model differs from zero, including its intercept. The centered-mean t test asks whether the fitted effect at a fixed weighted annotation reference differs from zero. A gene can pass an added test without showing annotation-related differences. Single-moderator and rare-AF models are secondary, and available-family gene-level BH values remain provisional because some source genes are unavailable.

The model and its notation

The F and t reference distributions use independent Gaussian residuals with sampling variances multiplied by a common gene-specific scale. These are working assumptions for variant summary estimates; LD and shared carriers can violate independence.

Baseline inclusion is AF≤0.05. Secondary fits use AF<0.001 and AF<0.0001. Native standard errors are used when supplied; otherwise they are approximated from the signed beta and two-sided variant p-value using the inverse normal tail. No MAC, variant-p, interval-QC or site-QC exclusions were imposed. Flagged and unknown-QC rows remain in the analysis.

For each quantitative annotation, missing values receive the gene’s observed baseline mean after transformation. Only genes with no observed values use the phenotype-wide observed fallback. Missing LoF receives the observed baseline HC 0/1 proportion, with the observed phenotype-wide fallback only when no baseline LoF labels are observed. These imputation means are unweighted observed-annotation means and stay fixed at the rare-AF cutoffs. The HC label is LOFTEE high confidence; a missing label is not automatically assigned zero. A fully imputed annotation is constant and cannot have a separately identifiable slope.




The centering reference c_g is different from the imputation means. It is frozen across AF cutoffs, and a rare-subset contrast is reported only if it lies in that subset’s estimable row space. Contrasts are evaluated in SVD coordinates to avoid cancellation with nearly constant annotations. At baseline, because the model includes an intercept and c_g uses the same inverse-variance weights, the centered estimate equals the baseline inverse-variance-weighted mean variant beta; its uncertainty still uses the full model’s residual scale. It is not a LoF burden estimate.
All-gene fitting used six spawn workers, worker-local read-only annotation SQLite and one BLAS thread per worker. Collection used one persistent eight-connection pool with a shared monotonic limit of four request starts per second. Documented failed requests remain unavailable sources, separate from valid empty responses.
Five primary annotation associations
| Gene | Variants | Annotation p | Planned Bonferroni p | Overall p | Centered p |
|---|---|---|---|---|---|
| MC4R | 253 | 3.56e-16 | 6.87e-12 | 8.57e-22 | 3.76e-10 |
| KIAA1109 | 2895 | 4.81e-09 | 9.27e-05 | 1.81e-08 | 0.656 |
| UBR2 | 1028 | 2.92e-07 | 0.00564 | 2.66e-07 | 0.0807 |
| ADARB1 | 484 | 3.5e-07 | 0.00675 | 6.28e-07 | 0.206 |
| DIDO1 | 2072 | 5.05e-07 | 0.00974 | 1.83e-06 | 0.909 |


MC4R · Appetite signaling with effects in both directions
MC4R encodes a central melanocortin receptor in energy-balance circuits. Human functional and genetic studies distinguish receptor loss of function, often associated with higher BMI, from gain-of-function alleles associated with lower BMI. V103I and I251L have experimental evidence for altered receptor signaling/trafficking. A single damage score cannot represent every molecular effect of this receptor. [source 1] [source 2]
The positive AlphaMissense contrast and positive HC-LoF contrast describe conditional differences among the variant-effect estimates. The joint annotation test can detect that structure even while protective and BMI-increasing alleles coexist.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 1.09 | [-1.2, 3.37] | 0.349 |
| AM | 0.165 | [0.122, 0.208] | 8.35e-13 |
| LoF | 0.397 | [0.193, 0.601] | 0.000158 |
This baseline model contains 253 variants, including21 observed HC-LoF labels. Missing annotations imputed: HMM 0, AlphaMissense 89, LoF 0. It retains0 flagged and 0 unknown interval-QC rows.
Among the displayed variants that most weaken the annotation evidence when removed: 18-60372169-C-T (p.Glu61Lys), beta +0.8625, AF6.35e-05, annotation p without it1.19e-13; 18-60372043-C-T (p.Val103Ile), beta-0.107, AF0.0201, annotation p without it2.94e-14; 18-60371598-ATC-A (p.Ile251TrpfsTer34), beta +0.7103, AF3.43e-05, annotation p without it6.52e-15. These are influence diagnostics, not a causal ranking.
Across all single-variant deletions,0 leave the annotation p-value above the planned-family threshold. The diagnostic fixes the original annotation imputation values and centering reference; it does not refit a new imputation policy for each deletion.
The plot also shows protective Val103Ile and Ile251Leu. Removing an allele can either weaken or strengthen a joint annotation association: Ile251Leu has a large opposing influence on this particular fitted annotation relationship, even though its negative BMI estimate is compatible with prior protective-allele evidence.
A local identical-summary sensitivity removes adjacent variants sharing beta, p-value and rounded allele count. Phase and shared carriers cannot be established from these summary data. The prespecified primary analysis retains all rows.
| Variants removed together | Span(bp) | Annotation p after cluster deletion |
|---|---|---|
18-60371975-A-T18-60371976-A-T | 1 | 1.83e-15 |
| Secondary AF restriction | Variants | Annotation p | Overall p | Centered p |
|---|---|---|---|---|
| AF_lt_0p001 | 251 | 2.96e-08 | 2.14e-13 | 0.176 |
| AF_lt_0p0001 | 240 | 2.31e-06 | 5.8e-15 | 0.0402 |

KIAA1109 · BLTP1/KIAA1109: a truncating-variant pattern with prior BMI evidence
The archived analysis symbol KIAA1109 corresponds to BLTP1, a bridge-like lipid transfer protein. Previous human exome studies implicated KIAA1109/BLTP1 in adult BMI; the present result therefore extends an existing genetic observation rather than establishing a previously unknown obesity gene. Its cellular functions provide context, but this regression does not identify the tissue or mechanism linking variation to adiposity. [source 1] [source 2]
The strongest conditional contrast is HC-LoF. A strong annotation association together with a weak centered-mean test means the evidence is about differences between variant classes, rather than a uniformly shifted mean across the weighted baseline variants.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.00326 | [-0.166, 0.16] | 0.969 |
| AM | 0.0029 | [-0.0042, 0.01] | 0.423 |
| LoF | 0.263 | [0.182, 0.344] | 1.86e-10 |
This baseline model contains 2,895 variants, including267 observed HC-LoF labels. Missing annotations imputed: HMM 320, AlphaMissense 1,020, LoF 0. It retains606 flagged and 0 unknown interval-QC rows.
Among the displayed variants that most weaken the annotation evidence when removed: 4-122353086-T-TG (p.Ala4679GlyfsTer15), beta +0.8675, AF1.4e-05, annotation p without it3.68e-08; 4-122238178-C-T (p.Gln1221Ter), beta +0.7756, AF1.14e-05, annotation p without it2.09e-08; 4-122246684-TA-T (p.Asn1730ThrfsTer21), beta +0.9643, AF8.89e-06, annotation p without it2.09e-08. These are influence diagnostics, not a causal ranking.
Across all single-variant deletions,0 leave the annotation p-value above the planned-family threshold. The diagnostic fixes the original annotation imputation values and centering reference; it does not refit a new imputation policy for each deletion.
A local identical-summary sensitivity removes adjacent variants sharing beta, p-value and rounded allele count. Phase and shared carriers cannot be established from these summary data. The prespecified primary analysis retains all rows.
| Variants removed together | Span(bp) | Annotation p after cluster deletion |
|---|---|---|
4-122349904-TG-T4-122349907-T-TC | 3 | 9.6e-09 |
4-122304881-A-T4-122304882-CT-C | 1 | 6.24e-09 |
4-122328276-CT-C4-122328280-T-C4-122328282-C-A | 6 | 6.06e-09 |
| Secondary AF restriction | Variants | Annotation p | Overall p | Centered p |
|---|---|---|---|---|
| AF_lt_0p001 | 2877 | 1.31e-07 | 3.3e-09 | 0.0276 |
| AF_lt_0p0001 | 2847 | 1.32e-07 | 6.85e-09 | 0.0607 |

UBR2 · Protein turnover and an independently reported human obesity association
UBR2 encodes an E3 ubiquitin-ligase component involved in protein turnover. A large human exome study reported obesity associations for rare protein-truncating UBR2 variants. That prior result supports the biological relevance of examining UBR2, but its published burden coefficient and the conditional annotation contrast estimated here answer different questions. [source 1]
The positive HC-LoF contrast is the clearest conditional moderator result. Individual positive and negative variant estimates coexist, so neither every truncating allele nor every UBR2 variant should be assigned one phenotypic direction from this gene-level result.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.248 | [-0.488, -0.00789] | 0.0429 |
| AM | 0.0248 | [0.00704, 0.0426] | 0.00625 |
| LoF | 0.416 | [0.245, 0.586] | 2e-06 |
This baseline model contains 1,028 variants, including72 observed HC-LoF labels. Missing annotations imputed: HMM 84, AlphaMissense 348, LoF 0. It retains105 flagged and 0 unknown interval-QC rows.
Among the displayed variants that most weaken the annotation evidence when removed: 6-42594302-G-T (p.Glu177Ter), beta +2.502, AF2.54e-06, annotation p without it1.45e-06; 6-42632642-T-C (p.Phe458Leu), beta +0.1659, AF0.000203, annotation p without it1.11e-06; 6-42617340-T-C, beta +1.999, AF2.54e-06, annotation p without it1.03e-06. These are influence diagnostics, not a causal ranking.
Across all single-variant deletions,0 leave the annotation p-value above the planned-family threshold. The diagnostic fixes the original annotation imputation values and centering reference; it does not refit a new imputation policy for each deletion.
A local identical-summary sensitivity removes adjacent variants sharing beta, p-value and rounded allele count. Phase and shared carriers cannot be established from these summary data. The prespecified primary analysis retains all rows.
| Variants removed together | Span(bp) | Annotation p after cluster deletion |
|---|---|---|
6-42573959-G-GTT6-42573965-A-T | 6 | 1.66e-06 |
6-42594288-A-G6-42594302-G-T | 14 | 1.17e-06 |
6-42679812-A-AGGCCGGGCGCGGTGGCTC6-42679813-A-ACGCC | 1 | 4.58e-07 |
| Secondary AF restriction | Variants | Annotation p | Overall p | Centered p |
|---|---|---|---|---|
| AF_lt_0p001 | 1024 | 1.75e-07 | 2.28e-07 | 0.931 |
| AF_lt_0p0001 | 1000 | 2.94e-07 | 4.32e-07 | 0.955 |

ADARB1 · RNA editing, feeding biology, and a cluster that needs scrutiny
ADARB1 encodes ADAR2, an enzyme that edits RNA. Transgenic mice misexpressing ADAR2 developed hyperphagia and obesity, including mice expressing an editing-deficient protein; this supplies experimental context without proving that the human alleles here act through the same mechanism. The regression is a candidate human annotation association, not a demonstration of a causal obesity mechanism. [source 1] [source 2]
Several conditional contrasts contribute, and the most influential local indels have identical effect estimates, P values and allele counts. Such repeated information is compatible with linked or complex alleles. Summary statistics cannot establish their phase or shared carriers, so a cluster-deletion sensitivity is shown alongside the unchanged primary test.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.202 | [0.0569, 0.347] | 0.00644 |
| AM | 0.0229 | [0.000397, 0.0454] | 0.0461 |
| LoF | 0.534 | [0.132, 0.937] | 0.00942 |
This baseline model contains 484 variants, including21 observed HC-LoF labels. Missing annotations imputed: HMM 0, AlphaMissense 189, LoF 0. It retains0 flagged and 0 unknown interval-QC rows.
Among the displayed variants that most weaken the annotation evidence when removed: 21-45185066-A-AGGGG (p.Met554ArgfsTer6), beta +1.263, AF2.54e-06, annotation p without it1.57e-06; 21-45185075-A-AAG (p.Ser557LysfsTer52), beta +1.263, AF2.54e-06, annotation p without it1.57e-06; 21-45185061-T-TG (p.Thr553HisfsTer6), beta +1.263, AF2.54e-06, annotation p without it1.56e-06. These are influence diagnostics, not a causal ranking.
Across all single-variant deletions,0 leave the annotation p-value above the planned-family threshold. The diagnostic fixes the original annotation imputation values and centering reference; it does not refit a new imputation policy for each deletion.
The nine records at chr21:45,185,054–45,185,079 share beta, p-value and rounded allele count. Removing this 25 bp cluster changes annotation p to 0.0455, despite each individual deletion retaining planned-family significance. Their phase and shared-carrier status are unavailable, so this result needs particular caution about repeated information or complex variant representation.
A local identical-summary sensitivity removes adjacent variants sharing beta, p-value and rounded allele count. Phase and shared carriers cannot be established from these summary data. The prespecified primary analysis retains all rows.
| Variants removed together | Span(bp) | Annotation p after cluster deletion |
|---|---|---|
21-45185054-C-CGGTAAG21-45185057-C-G21-45185061-T-TG21-45185063-ACC-A21-45185066-A-AGGGG21-45185068-GTC-G21-45185075-A-AAG21-45185077-T-G21-45185079-A-AG | 25 | 0.0455 |
21-45176169-G-A21-45176172-G-A | 3 | 3.78e-07 |
21-45176392-C-T21-45176445-G-C | 53 | 3.74e-07 |
| Secondary AF restriction | Variants | Annotation p | Overall p | Centered p |
|---|---|---|---|---|
| AF_lt_0p001 | 480 | 1.04e-06 | 3.24e-06 | 0.808 |
| AF_lt_0p0001 | 467 | 1.67e-06 | 1.46e-06 | 0.413 |

DIDO1 · Adipocyte biology with sensitivity to a few rare variants
Human exome work reported sex-dependent BMI associations for DIDO1 loss of function, especially in women. Experimental work separately linked DIDO to adipocyte differentiation: an N-terminal mutant mouse model had impaired adipogenesis and a lean phenotype. These differ in species, perturbation and study design; the mouse phenotype is not a direct prediction of the effect of heterozygous human truncations. The present both-sexes analysis cannot establish a sex interaction. [source 1] [source 2]
The positive conditional HC-LoF contrast accompanies a strong annotation F test. Its leave-one-variant-out profile is particularly important because a small number of rare variants can determine whether the planned threshold is crossed. That sensitivity affects the strength of interpretation without changing the prespecified primary fit.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.246 | [-0.00157, 0.494] | 0.0515 |
| AM | 0.0109 | [-0.00657, 0.0285] | 0.221 |
| LoF | 1.02 | [0.633, 1.41] | 2.48e-07 |
This baseline model contains 2,072 variants, including22 observed HC-LoF labels. Missing annotations imputed: HMM 21, AlphaMissense 758, LoF 0. It retains0 flagged and 0 unknown interval-QC rows.
Among the displayed variants that most weaken the annotation evidence when removed: 20-62896574-G-GA (p.Arg671SerfsTer18), beta +2.661, AF1.27e-06, annotation p without it3.53e-06; 20-62893945-GA-G (p.Ser941ProfsTer63), beta +1.497, AF2.59e-06, annotation p without it3.13e-06; 20-62906049-CTT-C (p.Lys475ArgfsTer15), beta +2.453, AF1.27e-06, annotation p without it2.95e-06. These are influence diagnostics, not a causal ranking.
Across all single-variant deletions,4 leave the annotation p-value above the planned-family threshold. The diagnostic fixes the original annotation imputation values and centering reference; it does not refit a new imputation policy for each deletion.
Four separate single-variant deletions cross the annotation-family threshold. Thus the original significant result depends on a small number of influential observations; it is not a robust claim that every DIDO1 truncation has the same BMI effect.
A local identical-summary sensitivity removes adjacent variants sharing beta, p-value and rounded allele count. Phase and shared carriers cannot be established from these summary data. The prespecified primary analysis retains all rows.
| Variants removed together | Span(bp) | Annotation p after cluster deletion |
|---|---|---|
20-62911223-T-A20-62911225-TG-T | 2 | 1.05e-06 |
20-62909742-G-A20-62909743-C-A | 1 | 5.15e-07 |
20-62879780-G-A20-62879781-G-A | 1 | 5.14e-07 |
| Secondary AF restriction | Variants | Annotation p | Overall p | Centered p |
|---|---|---|---|---|
| AF_lt_0p001 | 2056 | 9.86e-07 | 1.49e-06 | 0.269 |
| AF_lt_0p0001 | 2015 | 1.5e-06 | 2.47e-07 | 0.0319 |

What the added tests reveal
The union across the three per-test planned baseline families contains 47 genes. The 42 additional profiles below are significant in the overall or centered-mean test. They do not become primary annotation discoveries by passing an added test. Selecting a result from any of the three tests is assessed more conservatively against 57,861 hypotheses.

PCSK1, GIPR and SEC16B connect some added-test observations to established hormone or lipid biology. Other profiles remain statistical candidates requiring mechanism-specific follow-up. A significant weighted mean may be driven by a small number of relatively common, precisely estimated variants; the AF<0.001 and AF<0.0001 sensitivities are therefore essential. The gallery includes one variant figure for every additional significant gene, rather than implying that all of them are independently validated obesity genes.
TNKS1BP1 · Overall F / Centered-mean t
TNKS1BP1
Tankyrase 1 binding protein 1. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=1,451, rank=4. Annotation p=0.506, overall p=1.13e-16, centered-mean p=7.63e-19. Centered estimate-0.015596,95% CI[-0.019,-0.0122]. Conservative three-family adjusted p: overall6.53e-12; centered4.41e-14.
Imputed HMM:51; AlphaMissense:488; LoF:0/1,451; observed HC-LoF:65; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 3.21;0 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.119 | [-0.148, 0.386] | 0.381 |
| AM | 0.00142 | [-0.0173, 0.0201] | 0.881 |
| LoF | -0.101 | [-0.26, 0.058] | 0.213 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.338 | 0.808 |
| AF_lt_0p0001 | 0.652 | 0.587 |

NLRP1 · Overall F / Centered-mean t
NLRP1
NLRP1 links innate immune sensing to metabolic physiology in mouse experiments: loss of the inflammasome promoted adiposity, whereas activation altered IL-18-dependent metabolic responses. This provides experimental context, but the baseline centered association here does not show that the human variants cause obesity through that pathway. [source 1]
Baseline n=1,074, rank=4. Annotation p=0.852, overall p=5.47e-16, centered-mean p=1.88e-18. Centered estimate+0.011284,95% CI[0.0088,0.0138]. Conservative three-family adjusted p: overall3.16e-11; centered1.09e-13.
Imputed HMM:24; AlphaMissense:392; LoF:1/1,074; observed HC-LoF:64; flagged/unknown interval QC:61/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 2.19;0 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.0107 | [-0.137, 0.116] | 0.868 |
| AM | -0.00202 | [-0.0209, 0.0168] | 0.833 |
| LoF | -0.0467 | [-0.156, 0.0622] | 0.4 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.585 | 0.242 |
| AF_lt_0p0001 | 0.269 | 0.0337 |

MUC1 · Overall F / Centered-mean t
MUC1
MUC1 encodes a transmembrane mucin. A Taiwan Biobank study associated functional MUC1 variants with waist circumference and multiple cardiometabolic traits. The present profile is a weighted summary of BMI variant effects; it cannot distinguish a direct MUC1 mechanism from regional or correlated genetic effects. [source 1]
Baseline n=537, rank=3. Annotation p=0.408, overall p=6.61e-11, centered-mean p=3.05e-12. Centered estimate+0.034167,95% CI[0.0248,0.0436]. Conservative three-family adjusted p: overall3.83e-06; centered1.76e-07.
Imputed HMM:59; AlphaMissense:537; LoF:17/537; observed HC-LoF:57; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 8.23;1 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.14 | [-1.28, 1] | 0.81 |
| AM | not estimable | [not estimable, not estimable] | not estimable |
| LoF | -0.0623 | [-0.155, 0.0309] | 0.19 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.76 | 0.366 |
| AF_lt_0p0001 | 0.762 | 0.311 |

DCP1B · Overall F / Centered-mean t
DCP1B
Decapping mRNA 1B. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=385, rank=4. Annotation p=0.198, overall p=1.83e-10, centered-mean p=6.73e-12. Centered estimate+0.015979,95% CI[0.0115,0.0204]. Conservative three-family adjusted p: overall1.06e-05; centered3.89e-07.
Imputed HMM:8; AlphaMissense:127; LoF:0/385; observed HC-LoF:33; flagged/unknown interval QC:8/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 6.75;1 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.113 | [-0.109, 0.335] | 0.316 |
| AM | -0.0164 | [-0.0434, 0.0106] | 0.234 |
| LoF | 0.121 | [-0.0121, 0.253] | 0.0747 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.331 | 0.532 |
| AF_lt_0p0001 | 0.24 | 0.322 |

PCSK1 · Overall F / Centered-mean t
PCSK1
PCSK1 encodes prohormone convertase 1/3, connecting the result to hormone processing and established human obesity genetics. Human exome studies support effects of heterozygous loss-of-function variants. Here the added overall/centered tests, rather than the primary annotation test, cross the planned threshold. [source 1]
Baseline n=527, rank=4. Annotation p=0.000147, overall p=8.34e-12, centered-mean p=4.65e-10. Centered estimate+0.029626,95% CI[0.0205,0.0388]. Conservative three-family adjusted p: overall4.83e-07; centered2.69e-05.
Imputed HMM:61; AlphaMissense:185; LoF:0/527; observed HC-LoF:37; flagged/unknown interval QC:80/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 7.4;1 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.816 | [-0.196, 1.83] | 0.114 |
| AM | 0.00855 | [-0.00691, 0.024] | 0.278 |
| LoF | 0.367 | [0.179, 0.555] | 0.000139 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 1.97e-06 | 0.0935 |
| AF_lt_0p0001 | 1.66e-06 | 0.577 |

ANKRD11 · Overall F / Centered-mean t
ANKRD11
ANKRD11 is associated with neurodevelopmental and growth phenotypes in KBG syndrome. Published growth data emphasize stature and feeding-related variation, which matter when interpreting a height-derived measure such as BMI. The baseline association here does not establish that the same disease-causing alleles or mechanisms explain this BMI profile. [source 1]
Baseline n=2,951, rank=4. Annotation p=0.0523, overall p=2.66e-10, centered-mean p=5.46e-11. Centered estimate-0.012555,95% CI[-0.0163,-0.00882]. Conservative three-family adjusted p: overall1.54e-05; centered3.16e-06.
Imputed HMM:189; AlphaMissense:1,135; LoF:0/2,951; observed HC-LoF:19; flagged/unknown interval QC:38/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 2.78;0 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.242 | [0.0396, 0.445] | 0.0191 |
| AM | 0.00761 | [-0.00676, 0.022] | 0.299 |
| LoF | 0.089 | [-0.312, 0.489] | 0.663 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.00464 | 0.000637 |
| AF_lt_0p0001 | 0.0759 | 0.0103 |

CDH15 · Overall F / Centered-mean t
CDH15
Cadherin 15. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=973, rank=4. Annotation p=0.374, overall p=6.02e-09, centered-mean p=1.46e-10. Centered estimate-0.020468,95% CI[-0.0267,-0.0143]. Conservative three-family adjusted p: overall0.000349; centered8.45e-06.
Imputed HMM:7; AlphaMissense:374; LoF:0/973; observed HC-LoF:60; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 4.21;0 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.242 | [-0.846, 0.363] | 0.433 |
| AM | 0.0199 | [-0.0184, 0.0581] | 0.309 |
| LoF | -0.0723 | [-0.192, 0.047] | 0.235 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.199 | 0.654 |
| AF_lt_0p0001 | 0.29 | 0.262 |

CCDC171 · Overall F / Centered-mean t
CCDC171
Coiled-coil domain containing 171. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=1,232, rank=4. Annotation p=0.000151, overall p=1.14e-09, centered-mean p=1.36e-07. Centered estimate-0.015383,95% CI[-0.0211,-0.00969]. Conservative three-family adjusted p: overall6.58e-05; centered0.00789.
Imputed HMM:296; AlphaMissense:423; LoF:0/1,232; observed HC-LoF:128; flagged/unknown interval QC:302/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 4.88;2 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.585 | [-1.11, -0.0583] | 0.0295 |
| AM | -0.0334 | [-0.0509, -0.0158] | 0.000201 |
| LoF | 0.0199 | [-0.0403, 0.0801] | 0.516 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.644 | 0.298 |
| AF_lt_0p0001 | 0.909 | 0.556 |

LRP3 · Overall F / Centered-mean t
LRP3
LDL receptor related protein 3. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=825, rank=4. Annotation p=0.00703, overall p=1.84e-09, centered-mean p=3.4e-09. Centered estimate-0.014366,95% CI[-0.0191,-0.00965]. Conservative three-family adjusted p: overall0.000106; centered0.000197.
Imputed HMM:99; AlphaMissense:302; LoF:1/825; observed HC-LoF:38; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 2.74;0 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.221 | [-0.431, 0.874] | 0.505 |
| AM | -0.0185 | [-0.0314, -0.00567] | 0.00482 |
| LoF | -0.0769 | [-0.255, 0.101] | 0.397 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.536 | 0.201 |
| AF_lt_0p0001 | 0.307 | 0.0567 |

SEC16B · Overall F / Centered-mean t
SEC16B
SEC16B has prior BMI-locus evidence. Intestinal knockout experiments implicate chylomicron trafficking and lipid absorption, offering a specific metabolic context. This statistical association does not show that the analyzed human variants reproduce that tissue-specific perturbation. [source 1]
Baseline n=830, rank=4. Annotation p=0.118, overall p=3.17e-08, centered-mean p=3.38e-09. Centered estimate+0.025019,95% CI[0.0168,0.0332]. Conservative three-family adjusted p: overall0.00183; centered0.000196.
Imputed HMM:56; AlphaMissense:277; LoF:0/830; observed HC-LoF:66; flagged/unknown interval QC:31/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 7.63;2 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.297 | [-0.817, 0.223] | 0.262 |
| AM | 0.0228 | [-0.000608, 0.0462] | 0.0562 |
| LoF | -0.0469 | [-0.118, 0.0238] | 0.194 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.829 | 0.642 |
| AF_lt_0p0001 | 0.983 | 0.741 |

PABPC4 · Overall F / Centered-mean t
PABPC4
Poly(A) binding protein cytoplasmic 4. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=458, rank=4. Annotation p=0.215, overall p=8.59e-08, centered-mean p=4.72e-09. Centered estimate-0.03501,95% CI[-0.0465,-0.0235]. Conservative three-family adjusted p: overall0.00497; centered0.000273.
Imputed HMM:33; AlphaMissense:187; LoF:0/458; observed HC-LoF:21; flagged/unknown interval QC:23/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 8.32;1 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.046 | [-0.326, 0.418] | 0.808 |
| AM | -0.000708 | [-0.0363, 0.0349] | 0.969 |
| LoF | -0.264 | [-0.51, -0.0187] | 0.035 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.219 | 0.982 |
| AF_lt_0p0001 | 0.207 | 0.673 |

GABPB2 · Overall F / Centered-mean t
GABPB2
GA binding protein transcription factor subunit beta 2. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=301, rank=4. Annotation p=0.484, overall p=2.88e-07, centered-mean p=6.62e-09. Centered estimate+0.029596,95% CI[0.0198,0.0393]. Conservative three-family adjusted p: overall0.0167; centered0.000383.
Imputed HMM:11; AlphaMissense:106; LoF:0/301; observed HC-LoF:14; flagged/unknown interval QC:19/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 5.15;2 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -1.1 | [-2.53, 0.328] | 0.13 |
| AM | 0.00921 | [-0.0187, 0.0371] | 0.516 |
| LoF | -0.00649 | [-0.181, 0.168] | 0.942 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.706 | 0.61 |
| AF_lt_0p0001 | 0.493 | 0.712 |

DEPDC1 · Overall F / Centered-mean t
DEPDC1
DEP domain containing 1. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=526, rank=4. Annotation p=0.336, overall p=3.07e-07, centered-mean p=1.13e-08. Centered estimate+0.019931,95% CI[0.0132,0.0267]. Conservative three-family adjusted p: overall0.0178; centered0.000656.
Imputed HMM:133; AlphaMissense:159; LoF:1/526; observed HC-LoF:54; flagged/unknown interval QC:53/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 7.16;2 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.453 | [-0.98, 0.0733] | 0.0914 |
| AM | 0.000924 | [-0.0227, 0.0245] | 0.939 |
| LoF | 0.0146 | [-0.107, 0.136] | 0.813 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.142 | 0.445 |
| AF_lt_0p0001 | 0.125 | 0.393 |

CPNE1 · Overall F / Centered-mean t
CPNE1
Copine 1. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=425, rank=4. Annotation p=0.425, overall p=5.59e-07, centered-mean p=1.62e-08. Centered estimate-0.01943,95% CI[-0.0261,-0.0128]. Conservative three-family adjusted p: overall0.0324; centered0.000937.
Imputed HMM:30; AlphaMissense:170; LoF:32/425; observed HC-LoF:46; flagged/unknown interval QC:44/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 6.02;1 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.387 | [-0.31, 1.08] | 0.276 |
| AM | 0.00254 | [-0.0169, 0.022] | 0.797 |
| LoF | 0.0281 | [-0.0213, 0.0775] | 0.264 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.288 | 0.223 |
| AF_lt_0p0001 | 0.249 | 0.263 |

PRRC2A · Overall F / Centered-mean t
PRRC2A
Proline rich coiled-coil 2A. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=2,156, rank=4. Annotation p=0.622, overall p=8.82e-07, centered-mean p=1.64e-08. Centered estimate-0.011536,95% CI[-0.0155,-0.00754]. Conservative three-family adjusted p: overall0.051; centered0.00095.
Imputed HMM:46; AlphaMissense:681; LoF:0/2,156; observed HC-LoF:50; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 4.19;2 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.0966 | [-0.0636, 0.257] | 0.237 |
| AM | 0.00344 | [-0.00818, 0.0151] | 0.561 |
| LoF | -0.0116 | [-0.186, 0.163] | 0.896 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.802 | 0.83 |
| AF_lt_0p0001 | 0.545 | 0.145 |

EXOSC9 · Overall F / Centered-mean t
EXOSC9
Exosome component 9. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=345, rank=4. Annotation p=0.161, overall p=2.3e-07, centered-mean p=1.86e-08. Centered estimate+0.031575,95% CI[0.0208,0.0424]. Conservative three-family adjusted p: overall0.0133; centered0.00108.
Imputed HMM:82; AlphaMissense:119; LoF:0/345; observed HC-LoF:46; flagged/unknown interval QC:97/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 4.53;2 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.404 | [-0.225, 1.03] | 0.207 |
| AM | 0.0192 | [-0.0142, 0.0526] | 0.259 |
| LoF | -0.0819 | [-0.173, 0.00883] | 0.0767 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.133 | 0.757 |
| AF_lt_0p0001 | 0.19 | 0.523 |

NUTM2F · Overall F / Centered-mean t
NUTM2F
NUT family member 2F. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=550, rank=4. Annotation p=0.295, overall p=7.93e-07, centered-mean p=3.75e-08. Centered estimate-0.015646,95% CI[-0.0212,-0.0101]. Conservative three-family adjusted p: overall0.0459; centered0.00217.
Imputed HMM:78; AlphaMissense:187; LoF:0/550; observed HC-LoF:34; flagged/unknown interval QC:139/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 3.32;1 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.0117 | [-0.0239, 0.000573] | 0.0617 |
| AM | -0.00175 | [-0.0492, 0.0457] | 0.942 |
| LoF | -0.0369 | [-0.19, 0.116] | 0.637 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.335 | 0.0716 |
| AF_lt_0p0001 | 0.265 | 0.0727 |

SYT3 · Overall F / Centered-mean t
SYT3
Synaptotagmin 3. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=448, rank=4. Annotation p=0.0285, overall p=8.03e-08, centered-mean p=4.06e-08. Centered estimate+0.029688,95% CI[0.0192,0.0401]. Conservative three-family adjusted p: overall0.00464; centered0.00235.
Imputed HMM:14; AlphaMissense:178; LoF:0/448; observed HC-LoF:24; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 7.26;1 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.243 | [-0.12, 0.606] | 0.19 |
| AM | 0.033 | [0.00541, 0.0607] | 0.0192 |
| LoF | 0.144 | [-0.19, 0.478] | 0.398 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.335 | 0.834 |
| AF_lt_0p0001 | 0.179 | 0.455 |

SEMA3G · Overall F / Centered-mean t
SEMA3G
Semaphorin 3G. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=732, rank=4. Annotation p=0.00385, overall p=1e-07, centered-mean p=5.11e-07. Centered estimate-0.021733,95% CI[-0.0302,-0.0133]. Conservative three-family adjusted p: overall0.00581; centered0.0295.
Imputed HMM:52; AlphaMissense:244; LoF:0/732; observed HC-LoF:50; flagged/unknown interval QC:8/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 3.92;2 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.101 | [-0.235, 0.437] | 0.556 |
| AM | -0.0455 | [-0.0822, -0.00893] | 0.0148 |
| LoF | 0.129 | [0.0361, 0.223] | 0.00665 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.0263 | 0.0726 |
| AF_lt_0p0001 | 0.0345 | 0.407 |

PANK4 · Overall F / Centered-mean t
PANK4
Pantothenate kinase 4 (inactive). This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=779, rank=4. Annotation p=0.0374, overall p=2.42e-07, centered-mean p=1.12e-07. Centered estimate+0.025291,95% CI[0.016,0.0346]. Conservative three-family adjusted p: overall0.014; centered0.00646.
Imputed HMM:43; AlphaMissense:310; LoF:0/779; observed HC-LoF:29; flagged/unknown interval QC:53/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 3.83;3 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.0786 | [-0.754, 0.911] | 0.853 |
| AM | 0.0259 | [0.00553, 0.0463] | 0.0128 |
| LoF | 0.188 | [-0.0765, 0.453] | 0.163 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.057 | 0.0532 |
| AF_lt_0p0001 | 0.0186 | 0.00544 |

POM121C · Centered-mean t
POM121C
POM121 transmembrane nucleoporin C. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Recovered QC now supplies observed LoF labels and a rank-four fit. The final overall p=6.42×10⁻⁶ does not pass the planned overall threshold; its centered-mean test still passes. Earlier QC-incomplete overall values are superseded by this final model.
Baseline n=792, rank=4. Annotation p=0.7, overall p=6.42e-06, centered-mean p=1.22e-07. Centered estimate+0.016278,95% CI[0.0103,0.0223]. Conservative three-family adjusted p: overall0.371; centered0.00704.
Imputed HMM:115; AlphaMissense:302; LoF:0/792; observed HC-LoF:49; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 2.66;3 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.048 | [-0.0628, 0.159] | 0.395 |
| AM | 0.000804 | [-0.0106, 0.0122] | 0.89 |
| LoF | 0.0297 | [-0.0494, 0.109] | 0.461 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.0374 | 0.00328 |
| AF_lt_0p0001 | 0.338 | 0.0429 |

MTARC2 · Centered-mean t
MTARC2
Mitochondrial amidoxime reducing component 2. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=245, rank=4. Annotation p=0.724, overall p=1.21e-05, centered-mean p=2.35e-07. Centered estimate-0.04368,95% CI[-0.0598,-0.0275]. Conservative three-family adjusted p: overall0.698; centered0.0136.
Imputed HMM:46; AlphaMissense:80; LoF:0/245; observed HC-LoF:13; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 3;4 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.484 | [-0.724, 1.69] | 0.43 |
| AM | -0.027 | [-0.0831, 0.0292] | 0.345 |
| LoF | -0.0563 | [-0.448, 0.335] | 0.777 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.433 | 0.425 |
| AF_lt_0p0001 | 0.722 | 0.649 |

VPS13C · Overall F / Centered-mean t
VPS13C
Vacuolar protein sorting 13 homolog C. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=2,438, rank=4. Annotation p=0.048, overall p=6.43e-07, centered-mean p=2.67e-07. Centered estimate-0.012553,95% CI[-0.0173,-0.00778]. Conservative three-family adjusted p: overall0.0372; centered0.0155.
Imputed HMM:289; AlphaMissense:798; LoF:0/2,438; observed HC-LoF:208; flagged/unknown interval QC:240/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 2.44;3 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.22 | [0.052, 0.388] | 0.0103 |
| AM | 0.00854 | [-0.0108, 0.0278] | 0.386 |
| LoF | -0.0198 | [-0.0612, 0.0216] | 0.349 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.114 | 0.88 |
| AF_lt_0p0001 | 0.0626 | 0.113 |

COQ4 · Centered-mean t
COQ4
Coenzyme Q4. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=279, rank=4. Annotation p=0.43, overall p=8.08e-06, centered-mean p=2.96e-07. Centered estimate+0.020814,95% CI[0.013,0.0286]. Conservative three-family adjusted p: overall0.467; centered0.0171.
Imputed HMM:17; AlphaMissense:120; LoF:0/279; observed HC-LoF:29; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 3.94;4 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.222 | [-1.85, 2.29] | 0.833 |
| AM | 0.0107 | [-0.00205, 0.0235] | 0.0996 |
| LoF | 0.00316 | [-0.148, 0.154] | 0.967 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.664 | 0.757 |
| AF_lt_0p0001 | 0.709 | 0.952 |

DEF6 · Overall F / Centered-mean t
DEF6
DEF6 guanine nucleotide exchange factor. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=440, rank=4. Annotation p=0.0784, overall p=1.58e-06, centered-mean p=3.61e-07. Centered estimate+0.021249,95% CI[0.0132,0.0293]. Conservative three-family adjusted p: overall0.0914; centered0.0209.
Imputed HMM:17; AlphaMissense:170; LoF:0/440; observed HC-LoF:24; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 1.67;4 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.0179 | [-0.343, 0.307] | 0.914 |
| AM | 0.0231 | [-0.00494, 0.0512] | 0.106 |
| LoF | -0.317 | [-0.62, -0.0137] | 0.0406 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.0902 | 0.0487 |
| AF_lt_0p0001 | 0.255 | 0.342 |

OR10C1 · Overall F / Centered-mean t
OR10C1
Olfactory receptor family 10 subfamily C member 1. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=305, rank=3. Annotation p=0.0366, overall p=7.57e-07, centered-mean p=5.57e-07. Centered estimate+0.014017,95% CI[0.00863,0.0194]. Conservative three-family adjusted p: overall0.0438; centered0.0322.
Imputed HMM:3; AlphaMissense:104; LoF:0/305; observed HC-LoF:0; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 1.52;7 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.471 | [-1.56, 0.614] | 0.394 |
| AM | 0.0278 | [0.00555, 0.05] | 0.0145 |
| LoF | not estimable | [not estimable, not estimable] | not estimable |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.244 | 0.0458 |
| AF_lt_0p0001 | 0.486 | 0.136 |

DEPDC5 · Overall F / Centered-mean t
DEPDC5
DEP domain containing 5, GATOR1 subcomplex subunit. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=1,182, rank=4. Annotation p=0.034, overall p=9.83e-07, centered-mean p=6.07e-07. Centered estimate+0.011825,95% CI[0.0072,0.0164]. Conservative three-family adjusted p: overall0.0569; centered0.0351.
Imputed HMM:33; AlphaMissense:394; LoF:439/1,182; observed HC-LoF:40; flagged/unknown interval QC:61/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 1.58;5 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.0806 | [-0.0967, 0.258] | 0.373 |
| AM | 0.00806 | [-0.000623, 0.0167] | 0.0688 |
| LoF | -0.132 | [-0.245, -0.0196] | 0.0214 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.145 | 0.349 |
| AF_lt_0p0001 | 0.133 | 0.3 |

TRPM4 · Centered-mean t
TRPM4
Transient receptor potential cation channel subfamily M member 4. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=1,128, rank=4. Annotation p=0.5, overall p=2.1e-05, centered-mean p=7.31e-07. Centered estimate-0.014157,95% CI[-0.0197,-0.00858]. Conservative three-family adjusted p: overall1; centered0.0423.
Imputed HMM:20; AlphaMissense:444; LoF:1/1,128; observed HC-LoF:108; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 2.28;4 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.0462 | [-0.281, 0.374] | 0.782 |
| AM | 0.017 | [-0.00968, 0.0437] | 0.211 |
| LoF | 0.0142 | [-0.02, 0.0483] | 0.416 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.52 | 0.163 |
| AF_lt_0p0001 | 0.606 | 0.235 |

SLC22A31 · Centered-mean t
SLC22A31
Solute carrier family 22 member 31. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=396, rank=3. Annotation p=0.777, overall p=1.61e-05, centered-mean p=7.91e-07. Centered estimate-0.01964,95% CI[-0.0273,-0.0119]. Conservative three-family adjusted p: overall0.93; centered0.0458.
Imputed HMM:41; AlphaMissense:396; LoF:1/396; observed HC-LoF:20; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 3.17;2 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.289 | [-1.65, 2.23] | 0.769 |
| AM | not estimable | [not estimable, not estimable] | not estimable |
| LoF | 0.0236 | [-0.0527, 0.1] | 0.543 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.98 | 0.69 |
| AF_lt_0p0001 | 0.823 | 0.341 |

RAB21 · Centered-mean t
RAB21
RAB21 regulates endosomal and integrin trafficking. A large human coding-variant BMI study previously implicated RAB21, and intestinal Rab21 experiments support a role in epithelial maintenance and lipid handling. These observations provide biological context, not a demonstrated mechanism for the present centered-mean association; its rare-AF and case-deletion sensitivities remain essential. [source 1] [source 2] [source 3]
The leading influence is p.Ser224Phe: beta +0.0424, AF 0.0196. Removing it changes centered-mean p from 8.16×10⁻⁷ to 0.282; the AF<0.001 analysis removes this same variant and also has centered p=0.282. Ser224Phe was previously associated with BMI in the 2018 coding-variant study. This is a recovery of known coding-variant context, not a new rare-LoF or causal mechanism discovery.
Baseline n=153, rank=4. Annotation p=0.093, overall p=4.8e-06, centered-mean p=8.16e-07. Centered estimate+0.039187,95% CI[0.0241,0.0542]. Conservative three-family adjusted p: overall0.278; centered0.0472.
Imputed HMM:84; AlphaMissense:60; LoF:0/153; observed HC-LoF:14; flagged/unknown interval QC:59/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 5.54;2 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.692 | [-2.15, 0.763] | 0.349 |
| AM | -0.0415 | [-0.0994, 0.0164] | 0.158 |
| LoF | 0.247 | [0.0132, 0.48] | 0.0385 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.191 | 0.282 |
| AF_lt_0p0001 | 0.139 | 0.353 |

SLC35G5 · Centered-mean t
SLC35G5
Solute carrier family 35 member G5. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=415, rank=4. Annotation p=0.847, overall p=5.18e-05, centered-mean p=9.22e-07. Centered estimate+0.018101,95% CI[0.011,0.0252]. Conservative three-family adjusted p: overall1; centered0.0533.
Imputed HMM:0; AlphaMissense:149; LoF:0/415; observed HC-LoF:20; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 3.17;3 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.0475 | [-0.376, 0.471] | 0.826 |
| AM | 0.0201 | [-0.0827, 0.123] | 0.701 |
| LoF | -0.0624 | [-0.217, 0.0921] | 0.428 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.709 | 0.339 |
| AF_lt_0p0001 | 0.71 | 0.229 |

ELL · Centered-mean t
ELL
Elongation factor for RNA polymerase II. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=575, rank=4. Annotation p=0.186, overall p=1.06e-05, centered-mean p=1.14e-06. Centered estimate+0.026263,95% CI[0.0158,0.0368]. Conservative three-family adjusted p: overall0.611; centered0.0661.
Imputed HMM:76; AlphaMissense:195; LoF:0/575; observed HC-LoF:14; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 1.51;6 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.291 | [-1.61, 1.03] | 0.665 |
| AM | -0.0682 | [-0.13, -0.00652] | 0.0303 |
| LoF | 0.0834 | [-0.368, 0.535] | 0.717 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.225 | 0.28 |
| AF_lt_0p0001 | 0.659 | 0.212 |

NUP214 · Centered-mean t
NUP214
Nucleoporin 214. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=1,602, rank=4. Annotation p=0.166, overall p=9.7e-06, centered-mean p=1.25e-06. Centered estimate-0.016639,95% CI[-0.0233,-0.00993]. Conservative three-family adjusted p: overall0.561; centered0.072.
Imputed HMM:42; AlphaMissense:534; LoF:0/1,602; observed HC-LoF:97; flagged/unknown interval QC:140/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 3.35;2 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.289 | [0.0162, 0.562] | 0.0379 |
| AM | -0.00879 | [-0.0363, 0.0187] | 0.531 |
| LoF | -0.0299 | [-0.109, 0.0495] | 0.46 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.363 | 0.631 |
| AF_lt_0p0001 | 0.144 | 0.549 |

USP17L7 · Overall F
USP17L7
Ubiquitin specific peptidase 17 like family member 7. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=602, rank=4. Annotation p=0.00221, overall p=1.3e-06, centered-mean p=1.6e-05. Centered estimate+0.010271,95% CI[0.00563,0.0149]. Conservative three-family adjusted p: overall0.0754; centered0.924.
Imputed HMM:561; AlphaMissense:200; LoF:0/602; observed HC-LoF:39; flagged/unknown interval QC:602/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 1.5;8 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.24 | [0.0863, 0.393] | 0.00225 |
| AM | 0.0723 | [-0.0562, 0.201] | 0.27 |
| LoF | -0.12 | [-0.235, -0.00471] | 0.0414 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.00332 | 0.532 |
| AF_lt_0p0001 | 0.0074 | 0.211 |

ZNF101 · Centered-mean t
ZNF101
Zinc finger protein 101. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=309, rank=4. Annotation p=0.204, overall p=1.42e-05, centered-mean p=1.36e-06. Centered estimate+0.015101,95% CI[0.00907,0.0211]. Conservative three-family adjusted p: overall0.821; centered0.0784.
Imputed HMM:10; AlphaMissense:103; LoF:0/309; observed HC-LoF:40; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 5.42;3 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.167 | [-0.832, 0.498] | 0.621 |
| AM | -0.0073 | [-0.0713, 0.0567] | 0.823 |
| LoF | -0.0126 | [-0.0338, 0.00863] | 0.244 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.973 | 0.895 |
| AF_lt_0p0001 | 0.94 | 0.824 |

GPR151 · Centered-mean t
GPR151
GPR151 illustrates the need for restraint in target interpretation: initial rare-variant BMI associations were followed by human-knockout work that did not support a clinically meaningful BMI reduction from complete loss of function. A centered-mean association alone cannot settle that therapeutic question. [source 1]
Baseline n=354, rank=4. Annotation p=0.0654, overall p=4.9e-06, centered-mean p=1.51e-06. Centered estimate-0.044166,95% CI[-0.0619,-0.0264]. Conservative three-family adjusted p: overall0.284; centered0.0875.
Imputed HMM:5; AlphaMissense:136; LoF:1/354; observed HC-LoF:36; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 4.57;3 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.158 | [-0.545, 0.861] | 0.658 |
| AM | -0.0349 | [-0.0908, 0.021] | 0.22 |
| LoF | -0.0457 | [-0.0841, -0.00728] | 0.0199 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.625 | 0.495 |
| AF_lt_0p0001 | 0.176 | 0.38 |

C1orf167 · Centered-mean t
C1orf167
Chromosome 1 open reading frame 167. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=1,093, rank=3. Annotation p=0.205, overall p=1.03e-05, centered-mean p=1.88e-06. Centered estimate+0.014787,95% CI[0.00873,0.0208]. Conservative three-family adjusted p: overall0.598; centered0.109.
Imputed HMM:1,093; AlphaMissense:428; LoF:0/1,093; observed HC-LoF:144; flagged/unknown interval QC:3/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 5.21;7 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | not estimable | [not estimable, not estimable] | not estimable |
| AM | 0.00859 | [-0.00586, 0.023] | 0.244 |
| LoF | -0.00831 | [-0.0235, 0.00688] | 0.283 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.297 | 0.0754 |
| AF_lt_0p0001 | 0.0246 | 0.0973 |

C4B · Overall F
C4B
Complement C4B (Chido blood group). This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=265, rank=4. Annotation p=0.00195, overall p=1.94e-06, centered-mean p=2.27e-05. Centered estimate-0.023968,95% CI[-0.0349,-0.013]. Conservative three-family adjusted p: overall0.112; centered1.
Imputed HMM:95; AlphaMissense:102; LoF:0/265; observed HC-LoF:24; flagged/unknown interval QC:45/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 5.31;5 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | 0.0748 | [-0.132, 0.281] | 0.477 |
| AM | -0.0614 | [-0.138, 0.0154] | 0.117 |
| LoF | 0.0455 | [0.0203, 0.0706] | 0.000452 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.277 | 0.596 |
| AF_lt_0p0001 | 0.221 | 0.303 |

GIPR · Centered-mean t
GIPR
GIPR is an incretin receptor with experimental and human coding-variant evidence linking altered receptor function to BMI. The centered contrast here remains a weighted model summary; it does not establish the effect of receptor agonism or antagonism as a treatment. [source 1]
Baseline n=400, rank=4. Annotation p=0.468, overall p=5.01e-05, centered-mean p=2.08e-06. Centered estimate-0.039286,95% CI[-0.0553,-0.0233]. Conservative three-family adjusted p: overall1; centered0.12.
Imputed HMM:42; AlphaMissense:154; LoF:0/400; observed HC-LoF:41; flagged/unknown interval QC:94/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 2.23;12 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.985 | [-2.34, 0.371] | 0.154 |
| AM | -0.0103 | [-0.0439, 0.0232] | 0.546 |
| LoF | 0.0109 | [-0.0888, 0.11] | 0.831 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.0936 | 0.378 |
| AF_lt_0p0001 | 0.184 | 0.287 |

MST1R · Centered-mean t
MST1R
Macrophage stimulating 1 receptor. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=1,192, rank=4. Annotation p=0.592, overall p=7.26e-05, centered-mean p=2.32e-06. Centered estimate-0.022447,95% CI[-0.0317,-0.0132]. Conservative three-family adjusted p: overall1; centered0.134.
Imputed HMM:31; AlphaMissense:472; LoF:1/1,192; observed HC-LoF:105; flagged/unknown interval QC:0/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 1.75;59 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.0747 | [-0.372, 0.222] | 0.622 |
| AM | -0.01 | [-0.0325, 0.0125] | 0.383 |
| LoF | -0.0234 | [-0.0754, 0.0287] | 0.378 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.00333 | 0.000624 |
| AF_lt_0p0001 | 0.256 | 0.0273 |

CHD6 · Centered-mean t
CHD6
Chromodomain helicase DNA binding protein 6. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=1,705, rank=4. Annotation p=0.216, overall p=2.38e-05, centered-mean p=2.48e-06. Centered estimate+0.010845,95% CI[0.00634,0.0153]. Conservative three-family adjusted p: overall1; centered0.143.
Imputed HMM:58; AlphaMissense:568; LoF:0/1,705; observed HC-LoF:44; flagged/unknown interval QC:125/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 1.76;61 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.0179 | [-0.215, 0.179] | 0.858 |
| AM | -0.00283 | [-0.013, 0.00733] | 0.585 |
| LoF | 0.276 | [0.0115, 0.541] | 0.0408 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.185 | 0.711 |
| AF_lt_0p0001 | 0.175 | 0.763 |

NCAN · Overall F
NCAN
Neurocan. This profile supplies a model-based BMI association for follow-up; it does not establish a specific obesity mechanism or an intervention direction.
Baseline n=984, rank=4. Annotation p=0.00323, overall p=2.52e-06, centered-mean p=2.39e-05. Centered estimate+0.015286,95% CI[0.00822,0.0224]. Conservative three-family adjusted p: overall0.146; centered1.
Imputed HMM:73; AlphaMissense:329; LoF:0/984; observed HC-LoF:48; flagged/unknown interval QC:21/0. A constant or unidentifiable annotation is not evidence of a LoF effect. The largest single-variant removal reduces the selected test’s −log10(p) by 3.32;140 deletions cross back above its per-test threshold.
| Conditional term | Estimate | 95% t interval | Secondary t-test p |
|---|---|---|---|
| HMM | -0.184 | [-0.533, 0.166] | 0.303 |
| AM | 0.0168 | [-0.0178, 0.0514] | 0.342 |
| LoF | 0.285 | [0.124, 0.445] | 0.000531 |
| Secondary AF filter | Overall p | Centered p |
|---|---|---|
| AF_lt_0p001 | 0.00282 | 0.13 |
| AF_lt_0p0001 | 0.00376 | 0.415 |

Pathways and single-cell enrichment
We tested six prespecified selections: original annotation planned-family hits; overall p<10⁻⁴,10⁻⁵ and10⁻⁶; centered planned-family hits; and the union passing any of the three tests at 0.05/57,861. Each selection uses its own finite valid-test background. All calculations ran locally against frozen public references; no derived gene lists or backgrounds were sent to enrichment services.
Pathways use Reactome98 and WikiPathways20261110 with the verified HGNC-to-Ensembl mapping. Terms with 5–500 members in the relevant background are tested, including zero-overlap terms. BH correction covers both pathway databases together within each query. Single-cell analysis uses the Human Protein Atlas 25.1 published cell-type-enriched, group-enriched and enhanced gene assignments across 154 cell types; it tests all types with at least five markers in the field-specific testable background intersected with 18,400 genes measured in all 154 cell types. The atlas contains 20,162 genes in total, but 17,868 gene–cell measurements are absent. Those absent values remain missing; measured zeros are retained. This complete-measurement rule was recorded after a pre-test reference check and before computing any enrichment p-value. A separate sensitivity uses each cell type’s measured-gene background. This is marker overrepresentation using aggregated single-cell reference expression, not differential expression in BMI carriers.
| Gene selection | Reference | Query/background | Terms | Ordinary q<.05 | Matched q<.05 | Leading nominal term | p | BH q |
|---|---|---|---|---|---|---|---|---|
| annotation_planned | pathway | 5/18,703 | 3042 | 0 | 0 | Hunger and satiety | 0.00214 | 1 |
| annotation_planned | single_cell | 5/17,486 | 154 | 0 | 0 | retinal bipolar cells | 0.0112 | 1 |
| overall_1e4 | pathway | 88/18,736 | 3042 | 0 | 0 | Deadenylation-dependent mRNA decay | 0.0022 | 1 |
| overall_1e4 | single_cell | 87/17,498 | 154 | 0 | 0 | alveolar cells type 1 | 0.00886 | 0.532 |
| overall_1e5 | pathway | 43/18,736 | 3042 | 0 | 0 | Deadenylation-dependent mRNA decay | 0.000272 | 0.469 |
| overall_1e5 | single_cell | 42/17,498 | 154 | 0 | 0 | t-cells | 0.0154 | 0.738 |
| overall_1e6 | pathway | 27/18,736 | 3042 | 0 | 0 | Deadenylation-dependent mRNA decay | 6.66e-05 | 0.203 |
| overall_1e6 | single_cell | 26/17,498 | 154 | 0 | 0 | retinal bipolar cells | 0.0117 | 0.878 |
| centered_planned | pathway | 40/18,736 | 3042 | 0 | 0 | Deadenylation-dependent mRNA decay | 0.000219 | 0.403 |
| centered_planned | single_cell | 39/17,498 | 154 | 0 | 0 | alveolar cells type 1 | 0.00806 | 1 |
| any_three_conservative | pathway | 35/18,736 | 3042 | 0 | 0 | Deadenylation-dependent mRNA decay | 0.000147 | 0.446 |
| any_three_conservative | single_cell | 34/17,498 | 154 | 0 | 0 | retinal bipolar cells | 0.0292 | 1 |
Across the six queries, there are 0 query–pathway and 0 query–cell-type results with ordinary BH q<0.05. Each displayed leading term should be read with its corrected q-value. Nominal overlap is not evidence of pathway activation, a common direction of effect, or a causal cell type.
We also computed exact variant-count-decile-matched probabilities by convolving hypergeometric probability masses. The cell-type analysis adds a crossed matching sensitivity using mean-expression quintiles. Fixed-union and reference-covered backgrounds and a pooled six-query BH sensitivity are retained in the full results. Counts and ordinary probabilities were independently checked against Fisher’s exact test; leading probabilities against integer combinatorics; BH against an independent implementation; and matched probabilities against 100,000 stratified resamples for leading terms.



A null corrected enrichment result does not rule out hypothalamic or other cell-specific biology. The primary annotation query contains only five genes, and this atlas uses broad reference cell types rather than every neuronal subtype. Unavailable effect sources, reference coverage and limited query size constrain power.
What these results support—and what still needs testing
The primary annotation results support heterogeneity in variant BMI estimates related to the encoded annotations under the specified diagonal WLS model. The overall and centered tests identify additional full-model or mean associations. Their coefficients and means are conditional summaries of variant-effect estimates, not estimates obtained by regressing individual participants’ BMI on gene-carrier status.
Several practical limits matter. SEs reconstructed from p-values are approximate. LD, shared carriers and repeated information from complex alleles are not modeled by the diagonal covariance. Gene lists can share loci or variants and are not necessarily independent discoveries. Imputation uncertainty is not propagated. Pointwise confidence intervals and post-selection influence plots do not provide multiplicity-adjusted coverage. Unavailable sources can alter future results, so available-family gene BH remains provisional even after collection stops. Independent cohorts, genotype-level carrier/phase checks and targeted functional experiments are the appropriate next evidence, particularly for fragile or clustered signals.
Reproducibility and data access
The report provides the full19,287-row catalogue, distinct test fields, coefficients, imputation counts, rare-AF sensitivities and separate unavailable-source lists. The results archive includes executed code and provenance. Exact frozen raw and annotated gzip inputs are version-pinned separately and restored by checksum; no input values were truncated or re-encoded for delivery.
The accompanying reproducibility bundle contains the source equations, frozen enrichment definitions, query/background tables, variant-influence data, code and validation records. The source association statistics are available from Genebass, and all reference versions and checksums are recorded. Full frozen inputs and the complete all-gene report are retained with the analysis checkpoint.
References
- Genebass: Karczewski et al., Cell Genomics 2022
- MC4R gain-of-function: Lotta et al., Cell 2019
- Human exomes and obesity: Akbari et al., Science 2021
- Sex- and age-specific obesity genetics: Kaisinger et al., Cell Genomics 2023
- ADAR2 misexpression and hyperphagia: Singh et al., JBC 2007
- DIDO and adipogenesis: García-López et al., PNAS 2024
- AlphaMissense: Cheng et al., Science 2023
- Human single-cell atlas: Karlsson et al., Science Advances 2021
- Frozen Human Protein Atlas 25.1 data
- Reactome
- WikiPathways

Leave a Reply