High-throughput screening (HTS) approaches have enabled rapid routine testing of millions of compounds towards identifying novel “hit” molecules for therapeutic targets. However, many nuisance compounds, commonly labeled as PAINS, or Pan-Assay INterference compoundS, can interfere with the assay detection technology, leading to many false-positive hits. Thus, early detection of PAINS is of great interest and importance to the drug discovery community. Here, we have generated, integrated, and curated large libraries of HTS data for luciferase (firefly and nano) and thiol reactivity, all of which are common interference mechanisms. We have developed several Quantitative Structure-Interference Relationship (QSIR) models of nuisance behavior based on carefully curated HTS data generated to enable this study. All QSIR models were developed following best practices of QSAR modeling including data curation and external validation. The models showed a correct classification rate (CCR) for external datasets ranging between 66 and 80%. We selected 256 new compounds (128 hits and 128 random compounds) from the NIH Chemical Genomics Center library for experimental validation, and the respective external prediction accuracy was between 60-70%. We suggest that these validated QSIR models should be employed to identify nuisance compounds among the experimental hits instead of popular but limited and inaccurate alert-based approaches. These models can be applied both prospectively as part of chemical library design and retrospectively for the HTS hit triage. We have made both models and curated datasets publicly available at: https://github.com/alvesvm/hts_nuisance.

Computational prediction of nuisance artifacts among high-throughput screening hits

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.