Atomwise Blog

Model Fine-Tuning Improves Virtual High-Throughput Screening Performance

Written by Atomwise Inc. | Jul 27, 2020 11:30:00 PM

At the American Chemical Society Fall 2020 Virtual Meeting & Expo, several Atomwise members and partners were selected to present their research and work. Learn what our Atoms have been working on below and visit Atomwise at ACS Fall 2020 Virtual Meeting & Expo for other presentation sessions. 

Brad Worley, PhD 
Atomwise

Atomwise Collaborators: ...Jon Sorenson, PhD, Izhar Wallach, PhD

Title: Model Fine-Tuning for Enhanced Virtual High-Throughput Screening Performance

Division: COMP

 

Abstract

Structure-based convolutional neural networks like AtomNet® have proven invaluable in early-stage drug discovery efforts by rapidly searching very large virtual compound libraries for compounds having affinity for a given target protein. However, the question of how to “fine-tune” a network—using the results of a confirmatory experimental assay to enrich for active compounds in later rounds of virtual screening—remains open. This tuning is especially important when screening targets that are markedly different from any target previously observed by the network during training, a scenario known as “domain shift”. We demonstrate a simple continued training protocol that improves next-round enrichment with only ~100 assay data points. Results from a more advanced semi-supervised fine-tuning protocol designed to mitigate the effects of domain shift will also be presented.

 

Speaker Presentation

View on-demand video

Join our team

Our team is comprised of over 80 PhD scientists who contribute to a high-performance academic-like culture that fosters robust scientific and technical excellence. We strongly believe that data wins over opinions, and aim for as little dogma as possible in our decision making. Learn more about our team and opportunities at Atomwise.