SNPic: SNP Topic Modeling for Interpretable Representation Learning of Complex Phenotypes

Résumé fourni par la source

Here we introduce SNP Topic Modeling (SNPic), a representation learning framework that reframes GWAS summary statistics as structured biological corpora, where traits become documents and either co-occuring traits or associated genes become words, then decomposes these representations into latent, interpretable genetic topics using probabilistic topic models originally developed for natural language processing. The shift from effect-size algebra to representation learning redefines the analytical question: rather than asking ``how correlated are these traits?'', SNPic asks ``what latent biological topics compose each trait and which biological modules are shared across traits?''. This transforms pleiotropy analysis from a variance-partitioning problem into a structure-discovery problem.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.