Virtual compound screening using gene expression

This project leverages artificial intelligence and advanced machine learning to enable virtual compound screening that predicts how chemical structures will alter human gene expression, accelerating drug discovery and repurposing. Funded by a four-year, $1.7 million federal grant from the National Institutes of Health, the effort expands previous A.I.-based drug repurposing work by extending capabilities to novel and not-yet-synthesized compounds.

The core innovation is using machine learning models to predict a compound's gene expression profile directly from its chemical structure. Instead of the conventional, labor-intensive laboratory generation of gene expression signatures for millions of compounds, the computational approach allows researchers to rapidly search vast chemical spaces and identify candidates whose predicted expression profiles counteract disease-associated dysregulated genes. This global, expression-profile-focused strategy differs from many traditional drug discovery approaches that target single proteins; the team's aim is to restore broad patterns of dysregulated gene expression associated with disease states.

Computational screening narrows the search from millions or billions of potential molecules to a manageable set of high-potential candidates. Those compounds showing promising, disease-reversing predicted expression profiles will then be synthesized and moved into preclinical laboratory evaluation. The workflow couples in silico prediction with targeted synthesis and biological testing, allowing academic teams to address fundamental technical challenges and explore compounds that also carry potential for new intellectual property-facilitating eventual translation to patient therapies.

Initially, the team will design new compounds inspired by two drugs previously identified in earlier research as potential inhibitors of liver cancer and SARS-CoV-2. The project emphasizes the essential role of computer scientists in developing the advanced machine-learning methods required to search massive chemical spaces and make accurate expression-profile predictions. By integrating expertise across disciplines - including pharmacology, medicinal chemistry and biomedical informatics - the initiative seeks to overcome cost and feasibility barriers that preclude laboratory profiling at scale.

If successful, this A.I.-driven virtual screening platform could transform how researchers identify both repurposed and novel therapeutic candidates by enabling rapid, cost-effective prioritization of compounds based on predicted molecular impact on disease-related gene expression. The work embodies a strategic shift toward predictive, systems-level discovery that can accelerate the path from computational hypothesis to synthesized candidate and preclinical assessment.