A hidden network of gene relationships unifies behavioral and molecular research

Behavioral and molecular approaches often share topics of inquiry across basic and applied research fields. However, for historical and practical reasons, these approaches conceptualize scientific questions differently. This segregation is particularly damaging to translational fields that are in need of the combined strength of these two major approaches. While studies are increasingly recognizing the value of integrated approaches through hybrid methods, these efforts require great resources. To potentially relieve this interdisciplinary detachment, we uncover similarities between behavioral and molecular work in the publication network to test if the two approaches are more aligned than they appear, through deeper common concepts. To do this, we first create a model to predict genes that are relevant to behavioral publications from their abstracts. These predictions are in turn used to define gene-based similarity across the entire medical literature network. This model is built on-top of a Large Language Model (LLM), used to find patterns in the semantic structure of abstracts. Our experiments demonstrated the model’s efficacy in predicting gene annotations in molecular abstracts and showed the practical value of predictions made on behavioral research. We found similar increased relative risks for differentially expressed genes in disease-specific publications compared to the global publication population in both the molecular and behavioral groups. Additionally, the model was able to correlate gene predictions between molecular publications and their behavioral references. We instantiated this model in a free and easy to use search engine that finds related publications via the latent representations we’ve established. Using this model, Abstract2Gene facilitates novel collaborations by uncovering shared gene signatures across medical literature, broadening the reach of research to distant fields and promoting collaboration between molecular and behavioral researchers.