Date

Wednesday 15 Jul 2026

Time

15:15 - 15:45

Location

BW018

Supervisor

Anthe Janssen

2nd reviewer

Stephan Hacker

Jury

Anne Wentink

Antimicrobial resistance (AMR) is a well-known growing health problem. Methicillin-resistant Staphylococcus aureus (MRSA) infections rank among the six most lethal contributors to AMR-related deaths worldwide. Covalent inhibitors hold significant therapeutic potential, particularly in antibacterial drug discovery, as irreversible target engagement may reduce the likelihood of resistance development and as they can make new targets amenable for the development of antibiotics with novel modes-of-action. Virtual screening (VS) approaches are increasingly effective in accelerating drug discovery. However, their performance for covalent inhibitors is inferior to that for reversible inhibitors. This can, in part, be explained by the lack of standardized experimental benchmarking datasets for covalent inhibitors.

Covalent fragment screening has proven effective in accelerating drug discovery, resulting in a growing body of experimental screening datasets. Nevertheless, their application to VS validation is limited by the absence of resolved binding geometries, and a lack of standardized assay and statistical thresholds. Under carefully controlled and transparently reported experimental conditions, fragment-based drug discovery (FBDD) provides a promising foundation for the validation and optimization of VS methods, because of their large coverage of chemical space and (potential) high-throughput applications.

In this study, we developed a 192-well Gridded Regular Indexed Denaturating (192-GRID) Gel as highly parallelized competitive activity-based protein profiling (ABPP) platform. We envisioned that performing a screen with this platform on a biologically relevant target would yield a dataset that will be a highly valuable resource for the benchmarking of covalent virtual screening approaches. 2160 acrylamides were screened for binding to cysteine 112 from overexpressed 3-oxoacyl [acyl-carrier-protein] synthase 2 (FabH) from MRSA in bacterial lysate. The covalent fragment engagement was determined based on competition of FabH-specific probe TAMRA-LM015. Results were interpreted in a binary manner, with compounds showing ≤ 50% residual signal relative to DMSO controls classified as “hits”. The screening results showed a hit rate of ~5% with 86 binding (hits) and 1728 non-binding (non-hits) fragments. A hit verification experiment proved that 9 / 10 hits were retested as hits, showing excellent reproducibility of the method. The 192-GRID Gel is easily implemented using a simple 3D printer to print the casting system and is compatible with the Bio-Rad Sub-Cell GT Cell electrophoresis system, which will enable facile implementation of this method in other laboratories and for other target proteins.

This dataset was used to validate molecular docking scoring functions in a binary ‘yes / no’ binding manner. The results mainly show that the covalent docking scoring functions have very limited predictive power on target engagement for small covalent fragments under these conditions. Among the evaluated scoring functions, the ML-based RF-Score outperformed the native GOLD scoring functions. Notably, stronger correlations were observed between compound engagement and intrinsic chemical properties, such as warhead reactivity, lipophilicity, and molecular weight. Together, these findings suggest that these conventional docking scoring functions may insufficiently capture the outcome of covalent engagement in this assay context, but integrating machine-learning approaches and physicochemical descriptors may enhance virtual screening optimization for covalent inhibitors. This highlights the importance of high-quality data sets for the benchmarking of covalent docking scoring functions.