From mundane to surprising nonadditivity: drivers and impact on ML models.

Nonadditivity (NA) in Structure-Activity and Structure-Property Relationship (SAR) data is a rare but very information rich phenomenon. It can indicate conformational flexibility, structural rearrangements, and errors in assay results and structural assignment. While purely ligand-based conformational causes of NA are rather well understood and mundane, other factors are less so and cause surprising NA that has a huge influence on SAR analysis and ML model performance. We here report a systematic analysis across a wide range of properties (20 on-target biological activities and 4 physicochemical ADME-related properties) to understand the frequency of various different phenomena that may lead to NA. A set of novel descriptors were developed to characterize double transformation cycles and identify trends in NA. Double transformation cycles were classified into "surprising" and "mundane" categories, with the majority being classed as mundane. We also examined commonalities among surprising cycles, finding LogP differences to have the most significant impact on NA. A distinct behavior of NA for on-target sets compared to ADME sets was observed. Finally, we show that machine learning models struggle with highly nonadditive data, indicating that a better understanding of NA is an important future research direction.

Selecting the Reaction Path in On-Surface Synthesis through the Electron Chemical Potential in Graphene.

The organometallic on-surface synthesis of the eight-membered sp2 carbon-based ring cyclooctatetraene (C8H8, Cot) with the neighboring rare-earth elements ytterbium and thulium yields fundamentally different products for the two lanthanides, when conducted on graphene (Gr) close to the charge neutrality point. Sandwich-molecular YbCot wires of more than 500 Å length being composed of an alternating sequence of Yb atoms and upright-standing Cot molecules result from the on-surface synthesis with Yb. In contrast, repulsively interacting TmCot dots consisting of a single Cot molecule and a single Tm atom result from the on-surface synthesis with Tm. While the YbCot wires are bound through van der Waals interactions to the substrate, the dots are chemisorbed to Gr via the Tm atoms being more electropositive compared to Yb atoms. When the electron chemical potential in Gr is substantially raised (n-doping) through backside doping from an intercalation layer, the reaction product in the synthesis with Tm can be tuned to TmCot sandwich-molecular wires rather than TmCot dots. By use of density functional theory, it is found that the reduced electronegativity of Gr upon n-doping weakens the binding as well as the charge transfer between the reaction intermediate TmCot dot and Gr. Thus, the assembly of the TmCot dots to long TmCot sandwich-molecular wires becomes energetically favorable. It is thereby demonstrated that the electron chemical potential in Gr can be used as a control parameter in an organometallic on-surface synthesis to tune the outcome of a reaction.

A high quality, industrial data set for binding affinity prediction: performance comparison in different early drug discovery scenarios.

We release a new, high quality data set of 1162 PDE10A inhibitors with experimentally determined binding affinities together with 77 PDE10A X-ray co-crystal structures from a Roche legacy project. This data set is used to compare the performance of different 2D- and 3D-machine learning (ML) as well as empirical scoring functions for predicting binding affinities with high throughput. We simulate use cases that are relevant in the lead optimization phase of early drug discovery. ML methods perform well at interpolation, but poorly in extrapolation scenarios-which are most relevant to a real-world application. Moreover, we find that investing into the docking workflow for binding pose generation using multi-template docking is rewarded with an improved scoring performance. A combination of 2D-ML and 3D scoring using a modified piecewise linear potential shows best overall performance, combining information on the protein environment with learning from existing SAR data.

The Playbooks of Medicinal Chemistry Design Moves.

Large databases of biologically relevant molecules, such as ChEMBL, SureChEMBL, or compound collections of pharmaceutical or agrochemical companies, are invaluable sources of medicinal chemistry information, albeit implicit. We developed a modified matched molecular pair approach to systematically and exhaustively extract the transformations in these databases and distill them into snippets of explicit design knowledge that are easily interpretable and directly applicable. The resulting "playbooks of medicinal chemistry design moves" capture the collective pharmaceutical and agrochemical research expertise across multiple chemists, companies, targets, and projects. They can be queried in an automated fashion for systematic prospective design and compound generation. The ChEMBL playbook and an application to exploit it are available at https://github.com/mahendra-awale/medchem_moves.

Matched Molecular Series Analysis for ADME Property Prediction.

Generation and prioritization of new molecules are the most central part of the drug design process. Matched molecular series analysis (MMSA) has recently been proposed as a formal approach that captures both of these key elements of design. In order to better understand the power of MMSA and its specific limitations, we here evaluate its performance as an ADME property prediction tool. We use four large and diverse inhouse data sets, logD, microsomal clearance, CYP2C9, and CYP3A4 inhibition. MMSA follows the concept of parallel structure-activity relationship (SAR), where if two identical substituent series on different scaffolds show similarity in their property profiles, SAR from one series can be transferred to the other series. We test four different similarity metrics to identify pairs of molecular series where information can be transferred. We find that the best prediction performance is achieved by a combination of centered root-mean-square deviation (cRMSD) and a network score approach previously published by Keefer et al. However, cRMSD alone strikes the best balance between accuracy and the number of predictions that can be made. We identify statistical metrics that allow estimating when MMSA predictions will work, similar to the well-known applicability domain concept in machine learning. MMSA achieves a prediction accuracy that is comparable to a standard machine-learning model and matched molecular pair analysis. In contrast to machine learning, however, it is very easy to understand where MMSA predictions are coming from. Finally, to prospectively test the power of MMSA, we retested compounds that were strong outliers in the initial predictions and show how the MMSA model can help to identify erroneous data points.

Nonadditivity Analysis.

We introduce the statistics behind a novel type of SAR analysis named "nonadditivity analysis". On the basis of all pairs of matched pairs within a given data set, the approach analyzes whether the same transformations between related molecules have the same effect, i.e., whether they are additive. Assuming that the experimental uncertainty is normally distributed, the additivities can be analyzed with statistical rigor and sets of compounds can be found that show significant nonadditivity. Nonadditivity analysis can not only detect nonadditivity, potential SAR outliers, and sets of key compounds but also allow estimating an upper limit of the experimental uncertainty in the data set. We demonstrate how complex SAR features that inform medicinal chemistry can be found in large SAR data sets. Finally, we show how the upper limit of experimental uncertainty for a given biochemical assay can be estimated without the need for repeated measurements of the same protein-ligand system.

mmpdb: An Open-Source Matched Molecular Pair Platform for Large Multiproperty Data Sets.

Matched molecular pair analysis (MMPA) enables the automated and systematic compilation of medicinal chemistry rules from compound/property data sets. Here we present mmpdb, an open-source matched molecular pair (MMP) platform to create, compile, store, retrieve, and use MMP rules. mmpdb is suitable for the large data sets typically found in pharmaceutical and agrochemical companies and provides new algorithms for fragment canonicalization and stereochemistry handling. The platform is written in Python and based on the RDKit toolkit. It is freely available from https://github.com/rdkit/mmpdb .

Dilatation of Vascular Prostheses in Ascending Aortic Position: A Long-Term Follow-Up Computed Tomography Study with Comparison of Different Measurement Methods.

BACKGROUND: The aim of this study was to evaluate long-term dilatation of Hemashield Gold and Hemashield Platinum vascular prostheses in ascending aortic position using different measurement methods to obtain precise results. METHODS: Between 1999 and 2007, 73 patients with Stanford type A dissection received ascending aortic replacement with Hemashield Gold and Hemashield Platinum prostheses. Measurements were performed using multiplanar reconstruction mode of electrocardiogram (ECG)-gated, multislice spiral computed tomography (MSCT) in strictly orthogonal cross-sectional planes. Different methods of measurement were compared and maximum dilatation was estimated for different time spans. RESULTS: Diameters calculated from the measured circumference showed a significant (p = 0.037) but clinically not relevant difference (0.1 mm) to the mean between the largest and the shortest cross-sectional diameter of the prosthesis. Dilatation after 24.2 ± 10.2 months was 8.5 ± 4.5%. Long-term dilatation after 91.8 ± 34 months amounted to 11.8 ± 4.2%. CONCLUSION: Based on ECG-gated MSCT images, the presented methods of measurement provided reliable results. Long-term analysis shows low dilatation rates for Hemashield prostheses, which therefore can be considered as safe from this point of view. Nevertheless, a maximal dilatation of 20% could be relevant in valve sparing root replacement. It remains unclear if a dilatation like this contributes to the formation of suture aneurysms.

Matched Molecular Series: Measuring SAR Similarity.

Suggesting novel compounds to be made on the basis of similarity to a previously seen structure-activity relationship (SAR) requires a measure for SAR similarity. While SAR similarity has intuitively been used by medicinal chemists for decades, no systematic comparison of candidate similarity metrics has been published to date. With this publication, we attempt to close that gap by providing a statistical framework that allows comparison of SAR similarity metrics by their ability to rank series that provide the best activity prediction of novel substituents. This prediction is a result of a two-step process that involves (a) judging the similarity between series and (b) transferring the SAR from one series to the other. We tested several SAR similarity metrics and found that a centered RMSD (cRMSD) in combination with a lineaar-regression-based prediction interpolation ranks SAR profiles best. Based on that ranking we can, with a given confidence, suggest novel substituents to be tested. The superiority of the cRMSD can be explained on the basis of experimental uncertainty of affinity data and measured affinity differences. The ability to measure SAR similarity is central to applications like matched molecular series (MMS) analysis, whose applicability depends on whether there is a potential for SAR transferability between series. With the new SAR similarity metric introduced here, we show how MMS can be used in a medicinal chemistry setting as an idea generator and a semiquantitative prediction tool.

Binding Pose Flip Explained via Enthalpic and Entropic Contributions.

The anomalous binding modes of five highly similar fragments of TIE2 inhibitors, showing three distinct binding poses, are investigated. We report a quantitative rationalization for the changes in binding pose based on molecular dynamics simulations. We investigated five fragments in complex with the transforming growth factor ß receptor type 1 kinase domain. Analyses of these simulations using Grid Inhomogeneous Solvation Theory (GIST), pKA calculations, and a tool to investigate enthalpic differences upon binding unraveled the various thermodynamic contributions to the different binding modes. While one binding mode flip can be rationalized by steric repulsion, the second binding pose flip revealed a different protonation state for one of the ligands, leading to different enthalpic and entropic contributions to the binding free energy. One binding pose is stabilized by the displacement of entropically unfavored water molecules (binding pose determined by solvation entropy), ligands in the other binding pose are stabilized by strong enthalpic interactions, overcompensating the unfavorable water entropy in this pose (binding pose determined by enthalpic interactions). This analysis elucidates unprecedented details determining the flipping of the binding modes, which can elegantly explain the experimental findings for this system.

Identification of FAH domain-containing protein 1 (FAHD1) as oxaloacetate decarboxylase.

Fumarylacetoacetate hydrolase (FAH) domain-containing proteins occur in both prokaryotes and eukaryotes, where they carry out diverse enzymatic reactions, probably related to structural differences in their respective FAH domains; however, the precise relationship between structure of the FAH domain and the associated enzyme function remains elusive. In mammals, three FAH domain-containing proteins, FAHD1, FAHD2A, and FAHD2B, are known; however, their enzymatic function, if any, remains to be demonstrated. In bacteria, oxaloacetate is subject to enzymatic decarboxylation; however, oxaloacetate decarboxylases (ODx) were so far not identified in eukaryotes. Based on molecular modeling and subsequent biochemical investigations, we identified FAHD1 as a eukaryotic ODx enzyme. The results presented here indicate that dedicated oxaloacetate decarboxylases exist in eukaryotes.

Protease Inhibitors in View of Peptide Substrate Databases.

Protease substrate profiling has nowadays almost become a routine task for experimentalists, and the knowledge on protease peptide substrates is easily accessible via the MEROPS database. We present a shape-based virtual screening workflow using vROCS that applies the information about the specificity of the proteases to find new small-molecule inhibitors. Peptide substrate sequences for three to four substrate positions of each substrate from the MEROPS database were used to build the training set. Two-dimensional substrate sequences were converted to three-dimensional conformations through mutation of a template peptide substrate. The vROCS query was built from single amino acid queries for each substrate position considering the relative frequencies of the amino acids. The peptide-substrate-based shape-based virtual screening approach gives good performance for the four proteases thrombin, factor Xa, factor VIIa, and caspase-3 with the DUD-E data set. The results show that the method works for protease targets with different specificity profiles as well as for targets with different active-site mechanisms. As no structure of the target and no information on small-molecule inhibitors are required to use our approach, the method has significant advantages in comparison with conventional structure- and ligand-based methods.

Strong nonadditivity as a key structure-activity relationship feature: distinguishing structural changes from assay artifacts.

Nonadditivity in protein-ligand affinity data represents highly instructive structure-activity relationship (SAR) features that indicate structural changes and have the potential to guide rational drug design. At the same time, nonadditivity is a challenge for both basic SAR analysis as well as many ligand-based data analysis techniques such as Free-Wilson Analysis and Matched Molecular Pair analysis, since linear substituent contribution models inherently assume additivity and thus do not work in such cases. While structural causes for nonadditivity have been analyzed anecdotally, no systematic approaches to interpret and use nonadditivity prospectively have been developed yet. In this contribution, we lay the statistical framework for systematic analysis of nonadditivity in a SAR series. First, we develop a general metric to quantify nonadditivity. Then, we demonstrate the non-negligible impact of experimental uncertainty that creates apparent nonadditivity, and we introduce techniques to handle experimental uncertainty. Finally, we analyze public SAR data sets for strong nonadditivity and use recourse to the original publications and available X-ray structures to find structural explanations for the nonadditivity observed. We find that all cases of strong nonadditivity (ΔΔpKi and ΔΔpIC50 > 2.0 log units) with sufficient structural information to generate reasonable hypothesis involve changes in binding mode. With the appropriate statistical basis, nonadditivity analysis offers a variety of new attempts for various areas in computer-aided drug design, including the validation of scoring functions and free energy perturbation approaches, binding pocket classification, and novel features in SAR analysis tools.

Unique macrocycles in the Taiwan traditional Chinese medicine database.

Chemical space coverage within natural product databases is an important criterion for the selection of databases for virtual screening. Chemical space analysis can also provide valuable hints towards chemical substructures that are responsible for medical activity. We therefore developed a protocol for structurally characterizing the chemical space covered in specific natural product databases by comparing it to a "standard" natural product scaffold distribution. In this contribution, we analyzed the structural characteristics of the traditional Chinese medicine database@Taiwan as an example. While we did not find classes of very characteristic small molecule scaffolds, we found that there are a number of specific macrocyclic scaffolds that are highly enriched in the traditional Chinese medicine database@Taiwan and not documented elsewhere. This surprising finding points towards underused regions in chemical space with a big potential for biological impact.

Substrate-driven mapping of the degradome by comparison of sequence logos.

Sequence logos are frequently used to illustrate substrate preferences and specificity of proteases. Here, we employed the compiled substrates of the MEROPS database to introduce a novel metric for comparison of protease substrate preferences. The constructed similarity matrix of 62 proteases can be used to intuitively visualize similarities in protease substrate readout via principal component analysis and construction of protease specificity trees. Since our new metric is solely based on substrate data, we can engraft the protease tree including proteolytic enzymes of different evolutionary origin. Thereby, our analyses confirm pronounced overlaps in substrate recognition not only between proteases closely related on sequence basis but also between proteolytic enzymes of different evolutionary origin and catalytic type. To illustrate the applicability of our approach we analyze the distribution of targets of small molecules from the ChEMBL database in our substrate-based protease specificity trees. We observe a striking clustering of annotated targets in tree branches even though these grouped targets do not necessarily share similarity on protein sequence level. This highlights the value and applicability of knowledge acquired from peptide substrates in drug design of small molecules, e.g., for the prediction of off-target effects or drug repurposing. Consequently, our similarity metric allows to map the degradome and its associated drug target network via comparison of known substrate peptides. The substrate-driven view of protein-protein interfaces is not limited to the field of proteases but can be applied to any target class where a sufficient amount of known substrate data is available.

Geometric deep learning-guided Suzuki reaction conditions assessment for applications in medicinal chemistry.

Suzuki cross-coupling reactions are considered a valuable tool for constructing carbon-carbon bonds in small molecule drug discovery. However, the synthesis of chemical matter often represents a time-consuming and labour-intensive bottleneck. We demonstrate how machine learning methods trained on high-throughput experimentation (HTE) data can be leveraged to enable fast reaction condition selection for novel coupling partners. We show that the trained models support chemists in determining suitable catalyst-solvent-base combinations for individual transformations including an evaluation of the need for HTE screening. We introduce an algorithm for designing 96-well plates optimized towards reaction yields and discuss the model performance of zero- and few-shot machine learning. The best-performing machine learning model achieved a three-category classification accuracy of 76.3% (±0.2%) and an F 1-score for a binary classification of 79.1% (±0.9%). Validation on eight reactions revealed a receiver operating characteristic (ROC) curve (AUC) value of 0.82 (±0.07) for few-shot machine learning. On the other hand, zero-shot machine learning models achieved a mean ROC-AUC value of 0.63 (±0.16). This study positively advocates the application of few-shot machine learning-guided reaction condition selection for HTE campaigns in medicinal chemistry and highlights practical applications as well as challenges associated with zero-shot machine learning.

Synthesis of all four stereoisomers of 3-(tert-butoxycarbonyl)-3-azabicyclo[3.1.0]hexane-2-carboxylic acid.

A synthesis of all four stereoisomers of 3-(tert-butoxycarbonyl)-3-azabicyclo[3.1.0]hexane-2-carboxylic acid has been developed, thereby significantly shortening the known literature procedures for the syntheses of these unnatural amino acids. With a simple adjustment of the reaction conditions, we were able to obtain either pure cis or trans acid. Optical resolution was accomplished via diastereomeric salt formation or alternatively via chromatography on a chiral stationary phase. Finally, ab initio calculations gave an explanation for the observed cis selectivity in the initial step.

MM/GBSA binding energy prediction on the PDBbind data set: successes, failures, and directions for further improvement.

We validate an automated implementation of a combined Molecular Mechanics/Generalized Born Surface Area (MM/GBSA) method (VSGB 2.0 energy model) on a large and diverse selection of protein-ligand complexes (855 complexes). Although this data set is diverse with respect to both protein families and ligands, after carefully removing flawed structures, a significant correlation (R(2) = 0.63) between calculated and experimental binding affinities is obtained. Consistent explanations for "outlier" complexes are found. Visual analysis of the crystal structures and recourse to the original literature reveal that neglect of explicit solvent, ligand strain, and entropy contribute to the under- and overestimation of computed affinities. The limits of the Molecular Mechanics/Implicit Solvent approach to accurately estimate protein-ligand binding affinities is discussed as is the influence of the quality of protein-ligand complexes on computed free energy binding values.

Deriving static atomic multipoles from the electrostatic potential.

The description of molecular systems using multipolar electrostatics calls for automated methods to fit the necessary parameters. In this paper, we describe an open-source software package that allows fitting atomic multipoles (MTPs) from the ab initio electrostatic potential by adequate atom typing and judicious assignment of the local axis system. By enabling the simultaneous fit of several molecules and/or conformations, the package addresses issues of parameter transferability and lack of sampling for buried atoms. We illustrate the method by studying a series of small alcohol molecules, as well as various conformations of protonated butylamine.

Optimal force control of a permanent magnet linear synchronous motor with multiple shuttles.

Using permanent magnet linear synchronous machines for transportation tasks offers a higher flexibility in production plants compared to conventional conveyor solutions. In this context, passive transportation devices (shuttles) with permanent magnets are commonly used. When multiple shuttles are operated in close vicinity, disturbances due to magnetic interaction can occur. To allow for high-speed operation of the motor with high position control accuracy, these coupling effects must be considered. This paper presents a model-based control strategy that is based on a magnetic equivalent circuit model which is able to describe the nonlinear magnetic behavior at low computational costs. A framework is derived for the model calibration based on measurements. An optimal control scheme for the multi-shuttle operation is derived that allows to accurately track the desired tractive forces of the shuttles while minimizing the ohmic losses at the same time. The control concept is experimentally validated on a test bench and compared to a state-of-the-art field-oriented control concept typically used in industry.