Agentic AI Report Generation for the Potential Associations Between Understudied Genes and Liver Diseases
Abinanda Prabhakaran1, Daniel J. B. Clarke1, Daniel Zhu1, Avi Ma'ayan1
1Department of Pharmacological Sciences, Windreich Department of Artificial Intelligence and Human Health, Mount Sinai Center for Bioinformatics, Icahn School of Medicine at Mount Sinai, One Gustave L. Levy Place, Box 1603, New York, NY 10029 USAAbstract
A central focus of biomedical research is the discovery of disease-related genes. Such genes can harbor mutations that lead to disease manifestation, belong to a disease-related pathway, or serve as a biomarker for detecting the disease condition. The discovery of such disease-related genes suffers from literature biases whereby some genes are widely studied while many others are ignored with little knowledge about their function or disease association. Advent biotechnologies such as multiomics profiling coupled with Artificial Intelligence can be used to illuminate knowledge about understudied genes and their relation to human diseases. Here, we applied this approach to discover associations between under-studied genes and 14 primary human liver diseases. For each liver disease, we identified top understudied human genes that are most likely relevant to each disease using two ranking methods. In addition, for each liver disease, we identified studies from ARCHS4/RummaGEO that compared the disease tissue to healthy liver tissue. We processed the data from these studies into disease signatures and performed various downstream analyses to characterize these signatures. Next, for each liver disease, we created a report that describes the state of knowledge about the liver disease, the identified understudied genes, the results from the differential expression analysis, and a discussion about potential next steps that need to be taken to further validate the associations between the understudied genes and each disease. These reports are formatted into a report with figures, tables, and validated references. The report generation is assisted with a Large Language Model (LLM) that writes the text in the report. Overall, this approach can be extended to generate reports for over a thousand human diseases, and expanded to cover other related analyses in each report.