A machine learning model trained on a high-throughput antibacterial screen increases the hit rate of drug discovery

Abstract

Screening for novel antibacterial compounds in small molecule libraries has a low success rate. We applied machine learning (ML)-based virtual screening for antibacterial activity and evaluated its predictive power by experimental validation. We first binarized 29,537 compounds according to their growth inhibitory activity (hit rate 0.87%) against the antibiotic-resistant bacterium Burkholderia cenocepacia and described their molecular features with a directed-message passing neural network (D-MPNN). Then, we used the data to train an ML model that achieved a receiver operating characteristic (ROC) score of 0.823 on the test set. Finally, we predicted antibacterial activity in virtual libraries corresponding to 1,614 compounds from the Food and Drug Administration (FDA)-approved list and 224,205 natural products. Hit rates of 26% and 12%, respectively, were obtained when we tested the top-ranked predicted compounds for growth inhibitory activity against B. cenocepacia, which represents at least a 14-fold increase from the previous hit rate. In addition, more than 51% of the predicted antibacterial natural compounds inhibited ESKAPE pathogens showing that predictions expand beyond the organism-specific dataset to a broad range of bacteria. Overall, the developed ML approach can be used for compound prioritization before screening, increasing the typical hit rate of drug discovery.

Antibiofilm approaches as a new paradigm for treating infections

“…Indeed, ML has gained significant attention in recent years due to its ability to learn and adapt to complex data and make predictions with high accuracy, even in the absence of predefined features. ML algorithms can analyze large amounts of data and identify patterns and relationships that are not easily discernible to humans or other computational techniques (Rahman et al 2022), such as QSAR, virtual screening and molecular docking.…”

Pre-trained molecular representations enable antimicrobial discovery

“…The discovery of novel antimicrobial compounds is one area in which in-silico screening of large chemical libraries has been shown to be useful [35,20,32]. Instead of relying on a deep learning-based predictor, we aimed to investigate whether MolE's static representation could enable an easily accessible algorithm, such as XGBoost, to identify novel antimicrobial candidates.…”

“…Despite the success of sophisticated deep learning architectures for AMP prediction [26, 19, 36], their task specificity limits the transferability of the learned representation to novel tasks and molecule types. One seminal step toward a general-purpose approach is the use of Directed Message Passing Neural Networks (D-MPNNs) [45] in general antimicrobial discovery [35, 32, 22, 43]. Nonetheless, this approach involved the creation of a custom training set by in-house screening a large number of compounds (ranging from 2,000 to 39,000 molecules) for growth-inhibitory activity against each microbial species of interest.…”

ChemPrint: An AI-Driven Framework for Enhanced Drug Discovery

“…To date, no AI-discovered therapeutic has achieved FDA approval, and few have progressed beyond non-clinical animal model proof of concept. Despite advancements, current AI models struggle to improve hit rates sufficiently, as evidenced by Schrödinger’s pipeline-wide average in vitro hit rate of 26% [3], alongside other groups reporting rates of 12%, 14.2%, and 26% [4,5]. Of the compounds discovered to be hits few, if any, demonstrate significant chemical diversity.…”