Trait-Level Rather Than Individual-Level Resolution in Fitness Genomics: A Population-Calibrated Variant Prioritization Framework in a South Asian Cohort

Translating candidate gene panels into objective physiological trait interpretations remains a major hurdle in exercise genomics. Conventional total genotype scoring suffers from severe panel-composition bias, fails to evaluate unannotated or novel variants, and frequently fosters inflated claims regarding individual athletic prediction. Here, we present a population-calibrated, multi-tiered computational framework that evaluates genomic variation across six evidence streams integrating deep learning protein and genomic language models (ESM-2, Evo 2), missense and splice predictors (AlphaMissense, SpliceAI), and regional pathogenic-density metrics to generate a continuous Variant Priority Score (VPS; 1-10 scale). By dynamically renormalizing missing evidence streams rather than zero-filling them, the framework robustly scores unannotated variants lacking empirical database entries. Individual trait burden across 344 genes (spanning 17 physiological traits in Aerobic, Anaerobic, and Recovery domains) is aggregated and normalized against ancestry-matched expectations derived from gnomAD South Asian and an in-house reference cohort (N > 13,000). When applied to 960 South Indian individuals (584 male, 376 female; whole-exome sequencing), this normalization removed the dependence of expected trait burden on panel size by construction, and gene count explained only 2.7% of the variance in the resulting capacity scores (six traits interrogated by 16 to 33 genes). Variance partitioning of 4,764 individual-by-trait scores (six of the 17 traits: one aerobic, two anaerobic and three recovery) revealed that score variation was predominantly trait-driven (45.5%) rather than individual-driven (11.1%, p = 0.386), a pattern reproduced independently in males and females and with no statistically significant gender difference in capacity scores, indicating that, in this cohort, the candidate panel resolved differences between physiological traits but did not detect differences between individuals in capacity scores (no athletic phenotype was measured). Benchmarking against ClinVar classifications achieved an AUC of 0.74, which fell to 0.62 when archive-dependent streams were excluded, directly quantifying the role of shared data provenance. Replacing the in-house frequency reference with gnomAD South Asian shifted absolute scores (median fold change 0.73) but preserved trait ordering (Spearman 0.96), so scores are positions against a stated, population-specific reference. This framework provides an auditable, evidence-traceable baseline for trait-level research and functional variant prioritization while offering a critical methodological safeguard against over-interpreting unvalidated individual performance predictions.

Authors

Publication Details

Journal
medRxiv
Published
2026-11-12
DOI
https://doi.org/10.64898/2026.11.10.26367795
Primary Topic
Genetics and Physical Performance
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Trait-Level Rather Than Individual-Level Resolution in Fitness Genomics: A Population-Calibrated Variant Prioritization Framework in a South Asian Cohort

Hima Jyothi Challa, Prahashit Jayanthy, Kalyan Ram Uppaluri, KRISHNA KUMAR VEMPATI et al.
medRxiv
Genetics and Physical Performance
preprint

Trait-Level Rather Than Individual-Level Resolution in Fitness Genomics: A Population-Calibrated Variant Prioritization Framework in a South Asian Cohort

Hima Jyothi Challa, Prahashit Jayanthy, Kalyan Ram Uppaluri, KRISHNA KUMAR VEMPATI, Balaji Malluri, Madhu Bhanu Varma Rayala
preprint en

Abstract

Translating candidate gene panels into objective physiological trait interpretations remains a major hurdle in exercise genomics. Conventional total genotype scoring suffers from severe panel-composition bias, fails to evaluate unannotated or novel variants, and frequently fosters inflated claims regarding individual athletic prediction. Here, we present a population-calibrated, multi-tiered computational framework that evaluates genomic variation across six evidence streams integrating deep learning protein and genomic language models (ESM-2, Evo 2), missense and splice predictors (AlphaMissense, SpliceAI), and regional pathogenic-density metrics to generate a continuous Variant Priority Score (VPS; 1-10 scale). By dynamically renormalizing missing evidence streams rather than zero-filling them, the framework robustly scores unannotated variants lacking empirical database entries. Individual trait burden across 344 genes (spanning 17 physiological traits in Aerobic, Anaerobic, and Recovery domains) is aggregated and normalized against ancestry-matched expectations derived from gnomAD South Asian and an in-house reference cohort (N > 13,000). When applied to 960 South Indian individuals (584 male, 376 female; whole-exome sequencing), this normalization removed the dependence of expected trait burden on panel size by construction, and gene count explained only 2.7% of the variance in the resulting capacity scores (six traits interrogated by 16 to 33 genes). Variance partitioning of 4,764 individual-by-trait scores (six of the 17 traits: one aerobic, two anaerobic and three recovery) revealed that score variation was predominantly trait-driven (45.5%) rather than individual-driven (11.1%, p = 0.386), a pattern reproduced independently in males and females and with no statistically significant gender difference in capacity scores, indicating that, in this cohort, the candidate panel resolved differences between physiological traits but did not detect differences between individuals in capacity scores (no athletic phenotype was measured). Benchmarking against ClinVar classifications achieved an AUC of 0.74, which fell to 0.62 when archive-dependent streams were excluded, directly quantifying the role of shared data provenance. Replacing the in-house frequency reference with gnomAD South Asian shifted absolute scores (median fold change 0.73) but preserved trait ordering (Spearman 0.96), so scores are positions against a stated, population-specific reference. This framework provides an auditable, evidence-traceable baseline for trait-level research and functional variant prioritization while offering a critical methodological safeguard against over-interpreting unvalidated individual performance predictions.

medRxiv
Genetics and Physical Performance
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon