Trey E. Ideker

Medea, an AI agent, performs therapeutic reasoning across biological contexts by preserving context, verifying steps, and reconciling evidence, outperforming existing models in multiple domains.

HyperMap transfers perturbation responses across biological contexts using a meta-learning framework, achieving accurate predictions with fewer parameters than typical models.

The paper shows common and rare genetic variants converge on shared molecular networks for most human traits, influenced by heritability and tissue specificity.

scLong is a billion-parameter foundation model that captures long-range gene context in scRNA-seq by modeling all genes and integrating gene knowledge, outperforming existing models in various tasks.

The paper introduces MIRAGE, a multi-modal generative model that integrates protein sequence, interaction, and localization data to improve function prediction and module detection.

This study shows that DREAM complex activity influences somatic mutation burden, lifespan, and age-related disease, with reduced activity linked to longer lifespan and lower disease risk.

This paper identifies six PD subgroups based on cell-type-specific genetic networks and links NRG6 to reduced myelin integrity in oligodendrocytes.

This paper introduces MutationProjector, an AI model that converts tumor mutation profiles into cancer subtype representations, achieving strong performance in predicting treatment responses and classifying metastasis.

The paper shows that desmosome mutations in keratinocytes of the tumor microenvironment promote melanoma proliferation through reduced desmosome gene expression and increased melanoma cell growth.

The paper shows that somatic mutations correlate with methylation changes and can predict age similarly to epigenetic clocks.

The paper introduces FAIRSCAPE, a framework for packaging biomedical datasets with metadata to enhance AI-readiness, explainability, and FAIR compliance.

The paper evaluates 46 human interactomes to identify which are best for disease gene discovery and interaction prediction, finding large networks like HumanNet perform better for disease genes while smaller ones like DIP excel in prediction.

This paper introduces a Transformer model called G2PT to predict metabolic traits from genetic variants and shows it outperforms existing methods across populations.

ProteinAligner improves protein function and property prediction by integrating sequence, structure, and literature text modalities in a multi-modal pretraining framework.

DrugChat, a multi-modal LLM, predicts drug mechanisms and properties through free-form text, outperforming GPT-4 and specialized models in accuracy and versatility.

This study identifies 14 genome-wide significant loci for delay discounting in 134,935 individuals and shows genetic associations with 212 medical conditions.

ProteinChat is a multi-modal large language model that predicts protein functions from amino acid sequences, outperforming general-purpose LLMs and task-specific models.

This study identifies critical human host factors involved in SARS-CoV-2 multicycle replication through a genome-scale siRNA screen and validates potential antiviral targets.

The paper identifies shared molecular networks between common and rare variants associated with alcohol consumption, revealing genes involved in alcohol metabolism and comorbid psychiatric traits.

The paper uses in vitro evolution in haploid human cells to identify drug resistance mechanisms and validates them through gene expression experiments.

The paper introduces the CM4AI project, which aims to create AI-ready datasets of human cell architecture using multimodal data from cell lines.

The paper constructs a PPI network of 100 high-confidence ASD genes, identifies novel interactions, and uses AlphaFold-Multimer to prioritize variants for functional validation in organoids.

The paper uses SEUSS to analyze the effects of 115 RUNX1 mutations on RNA profiles in myelogenous leukemia cells, categorizing them into three functional groups.

NDEx IQuery provides a tool for network-based gene set analysis using diverse pathways and integrates with existing resources to enhance pathway interpretation.

The paper shows that yeast cells lacking Met15 can assimilate inorganic sulfur via Yll058w, but require exogenous methionine or hydrogen sulfide chelators to grow, challenging the traditional view of MET15 as essential for sulfur assimilation.

The paper integrates protein imaging and interaction data to create a unified map of eukaryotic cell structure, revealing new subcellular systems and protein associations.

The paper reveals unexpected complexity and multipotentiality of early heart fields through lineage and single-cell analyses, identifying new progenitor sources and developmental relationships.

The paper identifies genetic determinants of chemical resistance in yeast using in vitro evolution and whole genome sequencing, revealing key transcription factors involved in drug resistance.

This paper introduces a method to identify persistent structures in multiscale omics data, expanding cell type catalogs and revealing SARS-COV-2 protein interactions.

The paper identifies missing variants in the UK Biobank exome data due to incorrect read alignment and shows they can be recovered by adjusting alignment parameters.