Performance of machine-learning scoring functions in structure-based virtual screening

Abstract

Classical scoring functions have reached a plateau in their performance in virtual screening and binding affinity prediction. Recently, machine-learning scoring functions trained on protein-ligand complexes have shown great promise in small tailored studies. They have also raised controversy, specifically concerning model overfitting and applicability to novel targets. Here we provide a new ready-to-use scoring function (RF-Score-VS) trained on 15 426 active and 893 897 inactive molecules docked to a set of 102 targets. We use the full DUD-E data sets along with three docking tools, five classical and three machine-learning scoring functions for model building and performance assessment. Our results show RF-Score-VS can substantially improve virtual screening performance: RF-Score-VS top 1% provides 55.6% hit rate, whereas that of Vina only 16.2% (for smaller percent the difference is even more encouraging: RF-Score-VS top 0.1% achieves 88.6% hit rate for 27.5% using Vina). In addition, RF-Score-VS provides much better prediction of measured binding affinity than Vina (Pearson correlation of 0.56 and −0.18, respectively). Lastly, we test RF-Score-VS on an independent test set from the DEKOIS benchmark and observed comparable results. We provide full data sets to facilitate further research in this area (http://github.com/oddt/rfscorevs) as well as ready-to-use RF-Score-VS (http://github.com/oddt/rfscorevs_binary).

The impact of compound library size on the performance of scoring functions for structure-based virtual screening

“…Second, we do not tune the hyperparameters of the ML algorithm to training data in the case of ML-based SFs (we used RF with the default values of the scikit-learn implementation except for 500 as the number of trees in the forest and 100 as the number of features to consider when looking for the best split at each node). Third, as in previous studies (Wójcikowski et al, 2017;Yasuo and Sekijima, 2019;Chen et al, 2019), ML-based SFs are trained on data from the DUD-E benchmark and tested on a second benchmark: DEKOIS2.0 (Bauer et al, 2013). In this way, the exploitation of DUD-E bias by the ML algorithm is avoided, as DEKOIS2.0 actives and decoys were selected in a different manner than those in DUD-E.…”

“…When a 3D structure of the protein target is available and the binding site is known, this problem is more specifically called Structure-Based Virtual Screening (SBVS). An active is here a small organic molecule modulating the 3 was trained on complexes with a range of targets (Wójcikowski et al, 2017). Such generic SF can hence be applied to molecules complexed with any of these targets without retraining.…”

“…Unlike classical SFs, ML-based SFs provide more accurate binding affinity prediction owing to larger training sets (Li et al, 2019). The enrichment ability of ML-based SFs for SBVS also improve with more training instances (Li et al, 2011), with the largest training sets to date approaching one million instances (Wójcikowski et al, 2017). Big data approaches are also being used to construct massive test sets (Hoffmann and Gastreich, 2019).…”

Machine Learning Classification Models to Improve the Docking‐based Screening: A Case of PI3K‐Tankyrase Inhibitors

“…Instead they are trained on large datasets to model the relationships between affinity and various descriptors. According to the latest literature data, the machine learning scoring functions can estimate the binding energy most successfully.…”

Drug-target interaction prediction: A Bayesian ranking approach

“…Docking simulations are biologically well-accepted but time consuming and require 3D structures of targets. Furthermore, some researchers report that standard molecular docking scoring functions may be replaced by machine learning based scoring functions with improved prediction results [17,18].…”