Multimodal Language Models for Mapping the Genotype-Phenotype Relationship

Explore the complex relationship between genotypes and phenotypes through an advanced multimodal language model approach in this 57-minute conference talk. Delve into the integration of high-content genotyping methods like single-cell RNA sequencing with cutting-edge learning techniques to unravel the intricate connections between gene expression patterns and complex phenotypes. Learn about the development of a multimodal foundation model for analyzing human transcriptomics at the cellular level, offering enhanced resolution of cellular heterogeneity and improved phenotype annotation. Discover how this innovative approach uncovers potential cross-tissue biomarkers and investigates gene polyfunctionality, exemplified by the VWF gene in endothelial cells. Gain insights into the dynamic interplay between gene expression and phenotypic manifestation, and understand the potential of integrated genetics in revealing new dimensions of cellular function and complexity. The talk covers various topics, including the one-shot model, confusion matrix results, phenotype and genotype dynamics, polyfunctional genes, and genotype-phenotype relationships, concluding with a Q&A session.

- Intro + Background
- Aim1: One-shot model
- Results: Confusion Matrix
- Results: Phenotype Dynamics
- Results: Genotype Dynamics
- Results: Polyfunctional Genes
- Results: Genotype-Phenotype Relationships
- Q+A