In 1982, Kuntz et al. published an article with the title “A Geometric Approach to Macromolecule-Ligand Interactions”, where they described a method “to explore geometrically feasible alignment of ligands and receptors of known structure”. Since then, small molecule docking has been employed as a fast way to estimate the binding pose of a given compound within a specific target protein and also to predict binding affinity. Remarkably, the first docking method suggested by Kuntz and colleagues aimed to predict binding poses but very little was specified about binding affinity. This raises the question as to whether docking is the right tool to estimate binding affinity. The short answer is no, and this has been concluded in several comprehensive analyses. However, in this opinion paper we discuss several critical aspects that need to be reconsidered before a reliable binding affinity prediction through docking is realistic. These are not the only issues that need to be considered, but they are perhaps the most critical ones. We also consider that in spite of the huge efforts to enhance scoring functions, the accuracy of binding affinity predictions is perhaps only as good as it was 10–20 years ago. There are several underlying reasons for this poor performance and these are analyzed. In particular, we focus on the role of the solvent (water), the poor description of H-bonding and the lack of the systems’ true dynamics. We hope to provide readers with potential insights and tools to overcome the challenging issues related to binding affinity prediction via docking.

Treatment options for COVID‐19, caused by SARS‐CoV‐2, remain limited. Understanding viral pathogenesis at the molecular level is critical to develop effective therapy. Some recent studies have explored SARS‐CoV‐2–host interactomes and provided great resources for understanding viral replication. However, host proteins that functionally associate with SARS‐CoV‐2 are localized in the corresponding subnetwork within the comprehensive human interactome. Therefore, constructing a downstream network including all potential viral receptors, host cell proteases, and cofactors is necessary and should be used as an additional criterion for the validation of critical host machineries used for viral processing. This study applied both affinity purification mass spectrometry (AP‐MS) and the complementary proximity‐based labeling MS method (BioID‐MS) on 29 viral ORFs and 18 host proteins with potential roles in viral replication to map the interactions relevant to viral processing. The analysis yields a list of 693 hub proteins sharing interactions with both viral baits and host baits and revealed their biological significance for SARS‐CoV‐2. Those hub proteins then served as a rational resource for drug repurposing via a virtual screening approach. The overall process resulted in the suggested repurposing of 59 compounds for 15 protein targets. Furthermore, antiviral effects of some candidate drugs were observed in vitro validation using image‐based drug screen with infectious SARS‐CoV‐2. In addition, our results suggest that the antiviral activity of methotrexate could be associated with its inhibitory effect on specific protein–protein interactions.

Drug-target Interactions (DTIs) prediction plays a central role in drug discovery. Computational methods in DTIs prediction have gained more attention because carrying out in vitro and in vivo experiments on a large scale is costly and time-consuming. Machine learning methods, especially deep learning, are widely applied to DTIs prediction. In this study, the main goal is to provide a comprehensive overview of deep learning-based DTIs prediction approaches. Here, we investigate the existing approaches from multiple perspectives. We explore these approaches to find out which deep network architectures are utilized to extract features from drug compound and protein sequences. Also, the advantages and limitations of each architecture are analyzed and compared. Moreover, we explore the process of how to combine descriptors for drug and protein features. Likewise, a list of datasets that are commonly used in DTIs prediction is investigated. Finally, current challenges are discussed and a short future outlook of deep learning in DTI prediction is given.