GenoMorph improves disease prediction by reasoning across gene pathways
GenoMorph: Pathway-Grounded Genomic Disease Reasoning via Adaptive Latent Computation
Artificial Intelligence
Summary
Predicting diseases from genes is hard because many systems just remember known gene-disease pairs rather than understanding how genes work together in pathways. The authors created GenoMorph, a new system that looks at DNA sequences and figures out how genes interact in biological pathways to predict diseases. Unlike earlier methods, GenoMorph can reason through these interactions and works well even when gene names are hidden. This approach is more reliable and efficient, improving accuracy significantly on tough tests.
What this means in practice
- •For genomics software developers: Build disease prediction tools that rely on gene pathway interactions instead of memorized gene-disease links, improving robustness and prediction accuracy.
- •For biopharmaceutical data teams: Enhance drug target identification by analyzing genomic data through pathway-based reasoning to better understand disease mechanisms beyond direct gene associations.
- •For clinical bioinformatics teams: Develop diagnostic platforms that maintain accuracy even when standard molecular identifiers are missing by leveraging pathway-grounded genomic reasoning.$Commercial implications: Enables creation of more reliable clinical tools for genomic diagnosis that outperform existing gene association-based products.
Authors
Tanmoy Kanti Halder, Akash Ghosh, Arijit Roy, Sriparna Saha
Abstract
Large language models (LLMs) have demonstrated strong capabilities in biological reasoning; however, genomic disease inference remains largely dependent on memorized gene-disease associations rather than understanding biological pathways. This shortcut learning undermines robustness and generalization, and breaks down when molecular identifiers are unavailable. We present GenoMorph, a multimodal genomic reasoning framework that shifts disease prediction from associative gene-disease mapping toward pathway-grounded reasoning. GenoMorph couples a frozen DNA foundation model with question-conditioned cross-attention fusion, self-adaptive latent reasoning (LatentSp), a residual reasoning gate for iterative genomic evidence reinjection, and rejection sampling fine-tuning regularized by hierarchical optimal transport (OT). Rather than learning direct gene-disease mappings, GenoMorph aligns genomic sequence representations with latent pathway dynamics, enabling reasoning trajectories that follow molecular interactions before producing disease predictions. LatentSp dynamically allocates computation according to reasoning confidence, reducing unnecessary reasoning steps and improving inference efficiency. We further construct an anonymized benchmark from the Kyoto Encyclopedia of Genes and Genomes (KEGG), replacing every gene and molecular identifier with anonymous symbols while preserving sequences and pathway topology, thereby removing memorization shortcuts. GenoMorph raises the weighted F1 from 0.7863 (BioReason) to 0.9412, and rejection sampling fine-tuning with self-adaptive latent reasoning pushes it to 0.9725 while cutting latency nearly 60%. On the anonymized benchmark it reaches 0.9465 F1, substantially outperforming prior systems and confirming that accurate disease prediction can arise from pathway reasoning rather than memorized gene-disease associations.