LitGene: an interpretable transformer model integrating text and ontology for gene representation learning

LitGene is a new transformer model that combines scientific text and gene ontology data to improve gene representation learning. It performs well on protein property benchmarks and can predict gene functions and disease associations without prior examples. The model also offers interpretability tools and identified a gene linked to idiopathic pulmonary fibrosis, demonstrating its potential for generating biological hypotheses.