Dataset Information

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.

SUBMITTER: Rahman ASMZ

PROVIDER: S-EPMC9624395 | biostudies-literature | 2022 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

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

Rahman A S M Zisanur ASMZ Liu Chengyou C Sturm Hunter H Hogan Andrew M AM Davis Rebecca R Hu Pingzhao P Cardona Silvia T ST

PLoS computational biology 20221013 10

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-M ...[more]

PMID: 36228001

Similar Datasets

Project description:High-throughput screening (HTS), as one of the key techniques in drug discovery, is frequently used to identify promising drug candidates in a largely automated and cost-effective way. One of the necessary conditions for successful HTS campaigns is a large and diverse compound library, enabling hundreds of thousands of activity measurements per project. Such collections of data hold great promise for computational and experimental drug discovery efforts, especially when leveraged in combination with modern deep learning techniques, and can potentially lead to improved drug activity predictions and cheaper and more effective experimental design. However, existing collections of machine-learning-ready public datasets do not exploit the multiple data modalities present in real-world HTS projects. Thus, the largest fraction of experimental measurements, corresponding to hundreds of thousands of "noisy" activity values from primary screening, are effectively ignored in the majority of machine learning models of HTS data. To address these limitations, we introduce Multifidelity PubChem BioAssay (MF-PCBA), a curated collection of 60 datasets that includes two data modalities for each dataset, corresponding to primary and confirmatory screening, an aspect that we call multifidelity. Multifidelity data accurately reflect real-world HTS conventions and present a new, challenging task for machine learning: the integration of low- and high-fidelity measurements through molecular representation learning, taking into account the orders-of-magnitude difference in size between the primary and confirmatory screens. Here we detail the steps taken to assemble MF-PCBA in terms of data acquisition from PubChem and the filtering steps required to curate the raw data. We also provide an evaluation of a recent deep-learning-based method for multifidelity integration across the introduced datasets, demonstrating the benefit of leveraging all HTS modalities, and a discussion in terms of the roughness of the molecular activity landscape. In total, MF-PCBA contains over 16.6 million unique molecule-protein interactions. The datasets can be easily assembled by using the source code available at https://github.com/davidbuterez/mf-pcba.

Project description:BackgroundPraziquantel (PZQ) is the only widely available drug to treat schistosomiasis. Given the potential for drug resistance, it is prudent to search for novel therapeutics. Identification of anti-schistosomal chemicals has traditionally relied on phenotypic (whole organism) screening with adult worms in vitro and/or animal models of disease-tools that limit automation and throughput with modern microtiter plate-formatted compound libraries.MethodsA partially automated, three-component phenotypic screen workflow is presented that utilizes at its apex the schistosomular stage of the parasite adapted to a 96-well plate format with a throughput of 640 compounds per month. Hits that arise are subsequently screened in vitro against adult parasites and finally for efficacy in a murine model of disease. Two GO/NO GO criteria filters in the workflow prioritize hit compounds for tests in the animal disease model in accordance with a target drug profile that demands short-course oral therapy. The screen workflow was inaugurated with 2,160 chemically diverse natural and synthetic compounds, of which 821 are drugs already approved for human use. This affords a unique starting point to 'reposition' (re-profile) drugs as anti-schistosomals with potential savings in development timelines and costs.FindingsMultiple and dynamic phenotypes could be categorized for schistosomula and adults in vitro, and a diverse set of 'hit' drugs and chemistries were identified, including anti-schistosomals, anthelmintics, antibiotics, and neuromodulators. Of those hits prioritized for tests in the animal disease model, a number of leads were identified, one of which compares reasonably well with PZQ in significantly decreasing worm and egg burdens, and disease-associated pathology. Data arising from the three components of the screen are posted online as a community resource.ConclusionsTo accelerate the identification of novel anti-schistosomals, we have developed a partially automated screen workflow that interfaces schistosomula with microtiter plate-formatted compound libraries. The workflow has identified various compounds and drugs as hits in vitro and leads, with the prescribed oral efficacy, in vivo. Efforts to improve throughput, automation, and rigor of the screening workflow are ongoing.

Project description:BACKGROUND: One of the major issues in the fight against infectious diseases is the notable increase in multiple drug resistance in pathogenic species. For that reason, newly acquired high-throughput data on virulent microbial agents attract the attention of many researchers seeking potential new drug targets. Many approaches have been used to evaluate proteins from infectious pathogens, including, but not limited to, similarity analysis, reverse docking, statistical 3D structure analysis, machine learning, topological properties of interaction networks or a combination of the aforementioned methods. From a biological perspective, most essential proteins (knockout lethal for bacteria) or highly conserved proteins (broad spectrum activity) are potential drug targets. Ribosomal proteins comprise such an example. Many of them are well-known drug targets in bacteria. It is intuitive that we should learn from nature how to design good drugs. Firstly, known antibiotics are mainly originating from natural products of microorganisms targeting other microorganisms. Secondly, paleontological data suggests that antibiotics have been used by microorganisms for million years. Thus, we have hypothesized that good drug targets are evolutionary constrained and are subject of evolutionary selection. This means that mutations in such proteins are deleterious and removed by selection, which makes them less susceptible to random development of resistance. Analysis of the speed of evolution seems to be good approach to test this hypothesis. RESULTS: In this study we show that pN/pS ratio of genes coding for known drug targets is significantly lower than the genome average and also lower than that for essential genes identified by experimental methods. Similar results are observed in the case of dN/dS analysis. Both analyzes suggest that drug targets tend to evolve slowly and that the rate of evolution is a better predictor of drugability than essentiality. CONCLUSIONS: Evolutionary rate can be used to score and find potential drug targets. The results presented here may become a useful addition to a repertoire of drug target prediction methods. As a proof of concept, we analyzed GO enrichment among the slowest evolving genes. These may become the starting point in the search for antibiotics with a novel mechanism.

Project description:Micro- and nanoscale technologies have radically transformed biological research from genomics to tissue engineering, with the relative exception of microbial cell culture, which is still largely performed in microtiter plates and petri dishes. Here, we present nanoscale culture of the opportunistic fungal pathogen Candida albicans on a microarray platform. The microarray consists of 1,200 individual cultures of 30 nl of C. albicans biofilms ("nano-biofilms") encapsulated in an inert alginate matrix. We demonstrate that these nano-biofilms are similar to conventional macroscopic biofilms in their morphological, architectural, growth, and phenotypic characteristics. We also demonstrate that the nano-biofilm microarray is a robust and efficient tool for accelerating the drug discovery process: (i) combinatorial screening against a collection of 28 antifungal compounds in the presence of immunosuppressant FK506 (tacrolimus) identified six drugs that showed synergistic antifungal activity, and (ii) screening against the NCI challenge set small-molecule library identified three heretofore-unknown hits. This cell-based microarray platform allows for miniaturization of microbial cell culture and is fully compatible with other high-throughput screening technologies.Microorganisms are typically still grown in petri dishes, test tubes, and Erlenmeyer flasks in spite of the latest advances in miniaturization that have benefitted other allied research fields, including genomics and proteomics. Culturing microorganisms in small scale can be particularly valuable in cutting down time, cost, and reagent usage. This paper describes the development, characterization, and application of nanoscale culture of an opportunistic fungal pathogen, Candida albicans. Despite a more than 2,000-fold reduction in volume, the growth characteristics and drug response profiles obtained from the nanoscale cultures were comparable to the industry standards. The platform also enabled rapid identification of new drug candidates that were effective against C. albicans biofilms, which are a major cause of mortality in hospital-acquired infections.

Dataset Information

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

Publications

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

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets