I-Corps: Translation Potential of Early Drug Discovery Using Artificial Intelligence Meta-Modeling of Ligand-Protein Binding Affinities

This I-Corps project is focused on the development of artificial intelligence (AI)-driven strategies to accelerate early-stage drug discovery in broad therapeutic areas. Drug discovery and development is time consuming and costly with high failure rates. Success often depends on reliable identification of potential drug candidates in the early stage of the pipeline. Virtual screening of large numbers of compounds has been very useful in identifying additional drug candidates for experimental validation and subsequent optimization. However, there still exists an unmet need for improved virtual screening due to the billions of chemical compounds and unknown target molecules or biomarkers. AI technologies have had considerable impact in drug discovery. Improved predictions by AI-based technologies can significantly accelerate virtual screening or early-stage drug discovery and hence subsequent drug development in many disease areas, providing broad economic advantages in terms of time and cost in both biomedicine and healthcare. This I-Corps project utilizes experiential learning coupled with a first-hand investigation of the industry ecosystem to assess the translation potential of the technology. This solution is based on the development of a general meta-modeling framework of ligand-protein binding affinity prediction by integrating traditional physical docking tools and sequence-based artificial intelligence (AI) models. The technology has more than 1,000 pre-trained, sequence-based, deep learning models using 10 different architectures and more than 200 pre-trained machine-learning meta-models. The combined models have shown superior performance in three different benchmarks compared to exclusively structure-based AI models, suggesting that scalable virtual screening is possible without structural data for accurate prediction of binding affinities. A key advantage of the technology is to leverage the ensembling power of multiple tools and datasets in multi-dimensional ways, reducing model-specific bias and enhancing model-specific strengths. The technology expands the scope of diverse drug targets, providing new avenues for different diseases. In particular, the innovation may help discover and optimize small molecule ligands for challenging target proteins. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.