Bridging the genotype-phenotype gap with generative artificial intelligence

Le résumé fourni par la source

The genotype-phenotype gap is a persistent barrier to complex trait genetic dissection, worsened by the explosive growth of genomic data (1.5 billion variants identified in the UK Biobank WGS study) alongside persistently scarce and subjective human-defined phenotypes. Digital phenotyping offers a potential solution, yet existing tools fail to balance scalable non-manual phenotype generation and biological interpretability of these quantitative traits. Here we report AIPheno, the first generative AI-driven "phenotype sequencer" that bridges this gap. It enables high-throughput, unsupervised extraction of digital phenotypes from imaging data and unlocks their biological meaning via generative network analysis. AIPheno transforms imaging modalities into a rich source of quantitative traits, dramatically enhancing cross-species genetic discovery, including novel loci such as CCBE1 (humans), KITLG-TMTC3 (domestic pigeons), and SOD2-IGF2R (swine). Critically, its generative module decodes AI-derived phenotypes by synthesizing variant-specific images to yield actionable biological insights. For example, it clarifies how the OCA2-HERC2 locus pleiotropically links pigmentation to retinal vascular traits via vascular visibility modulation. Integrating scalable non-manual phenotyping, enhanced genetic discovery power, and generative mechanistic decoding, AIPheno establishes a transformative closed-loop paradigm. This work addresses the longstanding genotype-phenotype imbalance, redefines digital phenotype utility, and accelerates translation of genetic associations into actionable understanding with profound implications for human health and agriculture.

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