The study of structure-activity relationships (SARs) of small molecules is of fundamental importance in medicinal chemistry and drug design. Here, we introduce an approach that combines the analysis of similarity-based molecular networks and SAR index distributions to identify multiple SAR components present within sets of active compounds. Different compound classes produce molecular networks of distinct topology. Subsets of compounds related by different local SARs are often organized in small communities in networks annotated with potency information. Many local SAR communities are not isolated but connected by chemical bridges, i.e., similar molecules occurring in different local SAR contexts. The analysis makes it possible to relate local and global SAR features to each other and identify key compounds that are major determinants of SAR characteristics. In many instances, such compounds represent start and end points of chemical optimization pathways and aid in the selection of other candidates from their communities.

Introduction: Artificial intelligence (AI) has inspired computer-aided drug discovery. The widespread adoption of machine learning, in particular deep learning, in multiple scientific disciplines, and the advances in computing hardware and software, among other factors, continue to fuel this development. Much of the initial skepticism regarding applications of AI in pharmaceutical discovery has started to vanish, consequently benefitting medicinal chemistry. Areas covered: The current status of AI in chemoinformatics is reviewed. The topics discussed herein include quantitative structure-activity/property relationship and structure-based modeling, de novo molecular design, and chemical synthesis prediction. Advantages and limitations of current deep learning applications are highlighted, together with a perspective on next-generation AI for drug discovery. Expert opinion: Deep learning-based approaches have only begun to address some fundamental problems in drug discovery. Certain methodological advances, such as message-passing models, spatial-symmetry-preserving networks, hybrid de novo design, and other innovative machine learning paradigms, will likely become commonplace and help address some of the most challenging questions.Open data sharing and model development will play a central role in the advancement of drug discovery with AI.

In this contribution, the classification of protein binding sites using the physicochemical properties exposed to their pockets is presented. We recently introduced Cavbase, a method for describing and comparing protein binding pockets on the basis of the geometrical and physicochemical properties of their active sites. Here, we present algorithmic and methodological enhancements in the Cavbase property description and in the cavity comparison step. We give examples of the Cavbase similarity analysis detecting pronounced similarities in the binding sites of proteins unrelated in sequence. A similarity search using SARS M(pro) protease subpockets as queries retrieved ligands and ligand fragments accommodated in a physicochemical environment similar to that of the query. This allowed the characterization of the protease recognition pockets and the identification of molecular building blocks that can be incorporated into novel antiviral compounds. A cluster analysis procedure for the functional classification of binding pockets was implemented and calibrated using a diverse set of enzyme binding sites. Two relevant protein families, the alpha-carbonic anhydrases and the protein kinases, are used to demonstrate the scope of our cluster approach. We propose a relevant classification of both protein families, on the basis of the binding motifs in their active sites. The classification provides a new perspective on functional properties across a protein family and is able to highlight features important for potency and selectivity. Furthermore, this information can be used to identify possible cross-reactivities among proteins due to similarities in their binding sites.

We introduce the SAR matrix data structure that is designed to elucidate SAR patterns produced by groups of structurally related active compounds, which are extracted from large data sets. SAR matrices are systematically generated and sorted on the basis of SAR information content. Matrix generation is computationally efficient and enables processing of large compound sets. The matrix format is reminiscent of SAR tables, and SAR patterns revealed by different categories of matrices are easily interpretable. The structural organization underlying matrix formation is more flexible than standard R-group decomposition schemes. Hence, the resulting matrices capture SAR information in a comprehensive manner.

Here, we propose a new method (CLARITY; Clustering with Local shApe-based similaRITY) for the analysis of microarray time course experiments that uses a local shape-based similarity measure based on Spearman rank correlation. This measure does not require a normalization of the expression data and is comparably robust towards noise. It is also able to detect similar and even time-shifted sub-profiles. To this end, we implemented an approach motivated by the BLAST algorithm for sequence alignment. We used CLARITY to cluster the times series of gene expression data during the mitotic cell cycle of the yeast Saccharomyces cerevisiae. The obtained clusters were related to the MIPS functional classification to assess their biological significance. We found that several clusters were significantly enriched with genes that share similar or related functions.

Increasingly, drug-discovery processes focus on complete gene families. Tools for analyzing similarities and differences across protein families are important for the understanding of key functional features of proteins. Herein we present a method for classifying protein families on the basis of the properties of their active sites. We have developed Cavbase, a method for describing and comparing protein binding pockets, and show its application to the functional classification of the binding pockets of the protein family of protein kinases. A diverse set of kinase cavities is mutually compared and analyzed in terms of recurring functional recognition patterns in the active sites. We are able to propose a relevant classification based on the binding motifs in the active sites. The obtained classification provides a novel perspective on functional properties across protein space. The classification of the MAP and the c-Abl kinases is analyzed in detail, showing a clear separation of the respective kinase subfamilies. Remarkable cross-relations among protein kinases are detected, in contrast to sequence-based classifications, which are not able to detect these relations. Furthermore, our classification is able to highlight features important in the optimization of protein kinase inhibitors. Using small-molecule inhibition data we could rationalize cross-reactivities between unrelated kinases which become apparent in the structural comparison of their binding sites. This procedure helps in the identification of other possible kinase targets that behave similarly in "binding pocket space" to the kinase under consideration.

Graph neural networks are able to solve certain drug discovery tasks such as molecular property prediction and \textit{de novo} molecule generation. However, these models are considered 'black-box' and 'hard-to-debug'. This study aimed to improve modeling transparency for rational molecular design by applying the integrated gradients explainable artificial intelligence (XAI) approach for graph neural network models. Models were trained for predicting plasma protein binding, cardiac potassium channel inhibition, passive permeability, and cytochrome P450 inhibition. The proposed methodology highlighted molecular features and structural elements that are in agreement with known pharmacophore motifs, correctly identified property cliffs, and provided insights into unspecific ligand-target interactions. The developed XAI approach is fully open-sourced and can be used by practitioners to train new models on other clinically-relevant endpoints. File list (2) download file view on ChemRxiv jimenez2020color.pdf (5.85 MiB) download file view on ChemRxiv molgrad_series.csv (13.15 KiB)

Graphs are frequently used to describe the geometry and also the physicochemical composition of protein active sites. Here, the concept of graph alignment as a novel method for the structural analysis of protein binding pockets is presented. Using inexact graph-matching techniques, one is able to identify both conserved areas and regions of difference among different binding pockets. Thus, using multiple graph alignments, it is possible to characterize functional protein families and to examine differences among related protein families independent of sequence or fold homology. Optimized algorithms are described for the efficient calculation of multiple graph alignments for the analysis of physicochemical descriptors representing protein binding pockets. Additionally, it is shown how the calculated graph alignments can be analyzed to identify structural features that are characteristic for a given protein family and also features that are discriminative among related families. The methods are applied to a substantial high-quality subset of the PDB database and their ability to successfully characterize and classify 10 highly populated functional protein families is shown. Additionally, two related protein families from the group of serine proteases are examined and important structural differences are detected automatically and efficiently.