Bringing a new drug to market costs between $350 million to $4.4 billion and takes nearly 10 years. Pre-discovery and research alone require 5-6 years of manually testing millions of compounds and evaluating metabolization. This brute-force approach contributes to the escalating costs of clinical trials and the steep market prices of life-saving therapies.
​
WHAI's research aims to automate the research phase by targeting the top of the funnel—we propose a deep learning-powered pipeline that can process millions of initial compounds to identify the most promising candidates. The pipeline was piloted on Glutaminase-1 (GLS-1), an enzyme implicated in the progression of aggressive cancers, including glioblastoma, hepatocarcinoma, pancreatic cancer, bone cancer, and triple-negative breast cancer.
 
Using a hybrid deep learning approach with two advanced transformer-based chemical models (PubChem10, ChemBERTa), we distilled 10 million chEMBL compounds down to 100 potential inhibitors with strong predictive accuracy (RMSE: 0.601, R²: 0.786). Top candidates were evaluated using AutoDockVina to analyze their binding affinities toward the target and a comprehensive literature review validated the final selection of five viable molecular compounds.
​
This pipeline can reduce the time and costs of early drug discovery by providing a scalable infrastructure to target any enzyme. The lead compounds identified are strong candidates for experimental validation, and this research will be a step toward developing a GLS-1 inhibitor to halt tumor progression and prevent cancer from reaching advanced stages.