The Front Cover shows an emerging method for data mining ligand–target bioactivity matrices known as Chemogenomic Active Learning. As ligand–target pairs labeled as actives and inactives are systematically picked one-by-one, a set of decision trees serving as rules to explain active and inactive bioactivity is built. In this article, Rakers et al. show how the method can effectively model nuclear hormone receptor (NHR) and cytochrome P450 (CYP450) family-wide ligand–target interaction with only a fraction of available data. An actively learned model can be updated after new ligands are synthesized and assayed for activity, converging on hit discovery and optimization more effectively than brute-force screening. Cover artwork by Christin Rakers and J.B. Brown. More information can be found in the Full Paper by J.B. Brown et al. on page 511 in Issue 6, 2018 (DOI: 10.1002/cmdc.201700677).

The Cover Feature shows the different behavior of structural water molecules and bulk water. We have developed the AquaMMapS tool, which is able to identify stationary hydration sites close to the protein surface starting from molecular dynamics simulation. Root-mean-square fluctuation (i.e., a 1.4 Å cutoff) is employed to discriminate stationary from nonstationary water molecules. Space is organized into a grid, where cells crossed by stationary water molecules are selected and their occupancy computed along the simulation. Moreover, an empirical scoring function, the AquaMMapScore, has been developed to evaluate the penalty of a ligand displacing a stationary water hot spot. More information can be found in the Full Paper by Stefano Moro et al. on page 522 in Issue 6, 2018 (DOI: 10.1002/cmdc.201700564).

The Cover Feature shows three chipmunks involved in the creation, analysis, and clustering of the synthesizable virtual molecule library CHIPMUNK. Nearly 100 million compounds were generated with in silico reactions on accessible building blocks, and their descriptor profile was analysed. The library was clustered together with molecules from other public libraries in order to relate it to the known chemical space and to divide the huge library into manageable subsets. It serves as an idea generator and covers the chemical space beyond the rule of five and of protein–protein as well as protein–ligand interaction modulators. Artwork provided by Melanie Wilkesmann. More information can be found in the Full Paper by Oliver Koch et al. on page 532 in Issue 6, 2018 (DOI: 10.1002/cmdc.201700689).

Digital Drug Discovery: Guest Editors Andreas Bender (University of Cambridge) and Nathan Brown (BenevolentAI) present the 20 articles included in this Special Issue on Cheminformatics in Drug Design. As they summarize each article, they also discuss the common themes within in silico drug discovery that these papers represent.

The three databases of PubChem, ChemSpider, and UniChem capture the majority of open chemical structure records. Collectively, they constitute a massively enabling resource for cheminformatics, chemical biology, and drug discovery. It is important for users to have at least some appreciation of differential content to enable utility judgments for the tasks at hand. This turns out to be challenging. By comparing the three resources in detail, this review assesses their differences, some of which are not obvious.

Better in 3D: Exploration of activity cliffs in structure–activity relationship series is an excellent source of new insight. Three-dimensional matched molecular pairs (3D MMP) are an intuitive way to analyze existing data and to build models based on resulting observations. We present a concise summary of models and applications of recent years. Shown is a 3D MMP as published by Posy et al., J. Chem. Inf. Model. 2013, 53, 1576–1588 (reprinted with permission, © 2013 American Chemical Society).

Promiscuity cliff: Shown is an exemplary promiscuity cliff comprising two structural analogues that were extensively tested in screening assays and displayed an unexpectedly large difference in the number of targets (PD) they were active against. The compound on the right was consistently inactive in all assays.

Kinome-wide prediction of kinase inhibitors: The capabilities of proteochemometric (PCM) models to make large-scale predictions on the entire kinome was explored. The combination of a compound fingerprint with a protein fingerprint 1) improves the activity prediction for each kinase relative to individually trained models and 2) enables prediction of the activity of compounds for the entire kinome, including cancer-related resistance mutations.

Some truth behind false positives: Promiscuous inhibitors show up in enzyme assays as false positives, mainly because of aggregation. Molecular dynamics (MD) simulations were conducted for known aggregators and non-aggregators, revealing the forces and physicochemical properties behind the phenomenon of aggregation. This study also investigated the potential use of MD simulations as predictive tools in this regard.

Taming complexity: IChem is a suite of software dedicated to the analysis and comparison of three-dimensional molecular objects. It converts an intricate three-dimensional information into much simpler fingerprints or graphs, thereby enabling high-throughput comparisons and fueling machine learning models for predicting important features like protein–protein interfaces, druggable cavities, interaction patterns, and binding poses.

Water watch: The AquaMMapS tool was developed to individuate zones occupied by stationary water molecules during MD simulations. Individuation of stationary water voxels in a protein binding site can be exploited for drug design. AquaMMapScoring was developed to evaluate the penalty associated with different ligand substituents, considering electronegativity and capability of forming hydrogen bonds and stability of the water displaced.

A molecular atlas: A robust generative topographic map of fragment-like chemical space has been optimized. It accommodates more than 40 million molecules with no more than 17 heavy atoms, from the theoretically enumerated GDB-17 and real-world PubChem/ChEMBL databases. It serves as a library comparison tool to highlight biases in real-world molecules versus possible species from GDB-17. Specific patterns, proper to some libraries and absent from others, are highlighted.

Predictive model: A large dataset of K562 biological inhibitors is computationally treated to identify compounds that possibly have therapeutic action against β-thalassemia. A predictive computational model for K562 inhibition is developed and validated. The model facilitates fast and reliable virtual screening of new molecules and is freely available online.

In silico strategies: We used data curation, descriptor selection, machine learning algorithms, consensus modeling techniques, diverse validation strategies, and applicability domain analysis to develop quantitative structure–activity relationship (QSAR) models of compound plasma protein binding. Experimental data uncertainty was also assessed, helping us form reasonable expectations for potential models.

We describe the use of cross-classified multilevel modelling to analyse the results of similarity-based virtual screening searches using 2D fingerprints. We show that the choice of fingerprint is more important than the choice of similarity coefficient, and that multiple reference structures need to be used in benchmark studies such as this.

Finding the MCS (maximum common substructure) is a computationally difficult challenge, for which several attempts have been made to improve both the search speeds, and quality, of reported solutions. This article describes challenging benchmarks for a series of MCS algorithms, and reports the MCS type and algorithm combinations which generally yield the fastest and most sensible results in a chemoinformatic problem domain.

An MMP for MPs: A matched molecular pair analysis was used to examine the change in melting point (ΔMP) between molecules in a set of ∼275 000 compounds. We found many cases in which the ΔMP correlates with changes in functional groups and simple descriptors, such as number of hydrogen bond donors and acceptors. We observed that this method remains stable and scales well with larger datasets, indicating its utility as a simple privacy-preserving technique to analyze large proprietary databases and share findings.

Aiding lead optimisation: A data set of high-quality protein–ligand complexes spanning 121 protein targets was analysed for the presence of bioisosteric fragments. A pairwise analysis of all ligands for each target was carried out. The ligands were fragmented and a pair of fragments considered bioisosteric if they occupy a similar volume of the protein binding site. Only a small number of the bioisosteric pairs were found to be common to two or more targets.

Reusable structural cheminformatics for drug discovery: Data analytics platforms such as KNIME enable the creation of robust cheminformatics protocols that can easily be transferred, reused, and extended. 3D-e-Chem provides a set of KNIME nodes and reconfigurable workflows for data-driven drug discovery including: ligand-based target prediction, structure-based bioactivity data mapping and ligand scaffold replacement, and ligand repurposing.

Cavity filling: Desolvation and σ-hole/dispersion interactions are the driving forces between the small-molecule CD4 mimetics halogenated phenyl motif, called region 1, and residues F376–N377 in the HIV-1 gp120 envelope protein. In this study, we demonstrate that only quantum-mechanics-based methods can capture those non-standard effects, opening a new perspective in the rational optimization of region 1 to improve binding affinity.

Attention to the neglected: Chagas disease is a neglected tropical disease that affects more than eight million people in the Americas. Using a set of 1306 sesquiterpene lactones (SLs) obtained from SistematX, ligand- and structure-based virtual screening were performed. Afterward, these two methodologies were combined to select potential antichagasic SLs for the three parasitic forms of T. cruzi, establishing a possible mechanism of action.