Our hit identification method starts with a screening performance evaluation to select optimal docking software for the given protein target. During the screening process, we will comprehensively consider the consistency of the ligand binding pose and the binding score to improve the success rate of hit discovery. The workflow of our method is as follows:

1. We will use the compounds and crystal structure provided by the organizer to perform a docking-based virtual screening performance evaluation. The evaluation library will consist of 895 compounds and their corresponding decoys, which will be generated by the TocoDecoy method. We will employ common docking software such as AutoDock Vina, AutoDock-GPU, LeDock, PLANTS, UCSF DOCK 6 and Glide to screen the evaluation library. The enrichment factor will serve as the metrics to guide the selection of appropriate docking software for the target.
2. The two best-performing docking software will be selected to screen large libraries (i.e., Specs and ChemDiv). The maximum number of output poses will be limited to one for each software, resulting in up to two binding poses (one from each software) per docked molecule after screening. Molecules with two poses having docking scores better than the reference values (determined based on active compounds) will be selected. The heavy-atom RMSD between the two poses of each molecule will be calculated, and only the molecules with an RMSD of less than 2 angstroms will be kept, as they have better consistency of binding pose.
3. MM/GBSA method will be utilized to re-score the top 2000 ranked molecules retained from the previous step. Molecules with both poses having MM/GBSA scores better than the reference values (determined based on active compounds) will be picked, as they have better consistency of binding score. Finally, structural clustering and visual inspection will be performed to determine the 100 compounds for experimental testing.

Compared to other methods, our approach has the following advantages: (1) the docking software we use is selected based on performance evaluation and has better target specificity; (2) consistency validation of docking pose and docking score enables more accurate and reliable hierarchical screening; (3) the effectiveness of our screening strategy has been prevailed in multiple internal virtual screening projects, enabling us to successfully discover promising hits targeting USP7, HBV capsid protein, PGK1, ALDH2, etc.

Schrödinger

AutoDock Vina, AutoDock-GPU, LeDock, PLANTS, UCSF DOCK6, AmberTools, OpenBabel.

1. Zhe Wang, Hong Pan, Huiyong Sun, Yu Kang, Huanxiang Liu, Dongsheng Cao*, Tingjun Hou*, fastDRH: a webserver to predict and analyze protein-ligand complexes based on molecular docking and MM/PB(GB)SA computation, Briefings in Bioinformatics, 2022, bbac201.
2. Xujun Zhang, Chao Shen, Ben Liao, Dejun Jiang, Jike Wang, Zhenxing Wu, Hongyan Du, Tianyue Wang, Wenbo Huo, Lei Xu, Dongsheng Cao*, Chang-Yu Hsieh*, Tingjun Hou*, TocoDecoy: A New Approach to Design Unbiased Datasets for Training and Benchmarking Machine-Learning Scoring Functions, Journal of Medicinal Chemistry, 2022, 65, 7918-7932.
3. Zhe Wang, Huiyong Sun, Chao Shen, Xueping Hu, Junbo Gao, Dan Li, Dongsheng Cao*, Tingjun Hou*, Combined Strategies in Structure-based Virtual Screening, Physical Chemistry Chemical Physics, 2020, 22, 3149-3159.
4. Zhe Wang, Xuwen Wang, Yu Kang, HaiYang Zhong, Chao Shen, Xiaojun Yao, Dongsheng Cao*, Tingjun Hou*, Binding Affinity and Dissociation Pathway Predictions for a Series of USP7 Inhibitors with Pyrimidinone Scaffold by Multiple Computational Methods, Physical Chemistry Chemical Physics, 2020, 22, 5487-5499.
5. Zhe Wang, Xuwen Wang, Youyong Li, Tailong Lei, Ercheng Wang, Dan Li, Yu Kang, Feng Zhu, Tingjun Hou*, farPPI: a webserver for accurate prediction of protein-ligand binding structures for small-molecule PPI inhibitors by MM/PB(GB)SA methods, Bioinformatics, 2019, 35, 1777–1779.
6. Zhe Wang, Huiyong Sun, Xiaojun Yao, Dan Li, Lei Xu, Youyong Li, Sheng Tian, Tingjun Hou*, Comprehensive evaluation of ten docking programs on a diverse set of protein-ligand complexes: prediction accuracy of sampling power and scoring power, Physical Chemistry Chemical Physics, 2016, 18, 12964-12975.