Targeting Billion-Scale Compound Libraries with Deep Learning: A SARS-Cov-2 Example
Presenter
DescriptionDeep learning-based surrogate models can be effective approaches to speed up costly computations in large workflows. In this work, we present the evaluation of compounds from a large chemical database using surrogate DL models in place of the slower traditional docking approaches. Our work demonstrates significant improvement of the surrogate of over traditional approaches in speed, while maintaining an acceptable level of error. This workflow faced challenges in scaling the inference problem to reduce overall time in searching billions of compounds for potential COVID-19 inhibitors. We present different types of deep learning models that can be used as surrogates. We show the impact of using images, 2-D and 3-D drug descriptors in the feature set. We explore the sample size on which to train the models using two different approaches to construct the training set, and we illustrate the effect of different model optimization strategies. These models were subsequently used to search a much larger space of compounds and provided input into the selection of compounds for whole cell viral inhibition assays.
TimeWednesday, 7 July 202114:30 - 15:00 CEST
LocationHenry Dunant
Event Type
Minisymposium
CS and Math
Life Sciences