Nathaniel Thomas, Anya Iyer, Aneri Sheth, Zachary Bashkin, and Robert Downing
With advancements in the field of machine learning-powered by on-demand computing and information processing on a large scale, computationally driven proteomics and high throughput virtual screening have become increasingly popular in reducing traditional in vitro screening costs and the timeframe for hit-to-lead identification of drug candidates. The efficiency of high throughput fingerprinting using cheminformatics-based approaches coupled with machine learning holds immense potential in screening possible inhibitors. To identify these potential targets, we propose a reductionist approach in identifying key pharmacophoric elements of chemical entities, dramatically reducing the relative compute cost for large-scale chemical screening efforts. By minimizing the 3D structure of our molecules to their key points we are able to screen a larger sample of chemical space while effectively filtering for ideal small molecule drugs. Platforms such as PaDEL and Mordred were used in identifying notable descriptors of a class of FDA-approved NNRTIs, and this data was later implemented in a machine learning-based model when screening for the structural similarity between NNRTIs and other datasets of organic compounds. Herein, we present a novel pharmacophore fingerprinting method based on 3D reductions of molecule libraries, enabling a relatively more efficient and rapid screening of effective inhibitors of the HIV-1 RT enzyme.