Task-Similarity is a Crucial Factor for Few-Shot Meta-Learning of Structure-Activity Relationships.

Machine learning models support computer-aided molecular design and compound optimization. However, the initial phases of drug discovery often face a scarcity of training data for these models. Meta-learning has emerged as a potentially promising strategy, harnessing the wealth of structure-activity data available for known targets to facilitate efficient few-shot model training for the specific target of interest. In this study, we assessed the effectiveness of two different meta-learning methods, namely model-agnostic meta-learning (MAML) and adaptive deep kernel fitting (ADKF), specifically in the regression setting. We investigated how factors such as dataset size and the similarity of training tasks impact predictability. The results indicate that ADKF significantly outperformed both MAML and a single-task baseline model on the inhibition data. However, the performance of ADKF varied across different test tasks. Our findings suggest that considerable enhancements in performance can be anticipated primarily when the task of interest is similar to the tasks incorporated in the meta-learning process.

Integrating Reaction Schemes, Reagent Databases, and Virtual Libraries into Fragment-Based Design by Reinforcement Learning.

Lead optimization supported by artificial intelligence (AI)-based generative models has become increasingly important in drug design. Success factors are reagent availability, novelty, and the optimization of multiple properties. Directed fragment-replacement is particularly attractive, as it mimics medicinal chemistry tactics. Here, we present variations of fragment-based reinforcement learning using an actor-critic model. Novel features include freezing fragments and using reagents as the fragment source. Splitting molecules according to reaction schemes improves synthesizability, while tuning network output probabilities allows us to balance novelty versus diversity. Combining fragment-based optimization with virtual library encodings allows the exploration of large chemical spaces with synthesizable ideas. Collectively, these enhancements influence design toward high-quality molecules with favorable profiles. A validation study using 15 pharmaceutically relevant targets reveals that novel structures are obtained for most cases, which are identical or related to independent validation sets for each target. Hence, these modifications significantly increase the value of fragment-based reinforcement learning for drug design. The code is available on GitHub: https://github.com/Sanofi-Public/IDD-papers-fragrl.

Interpretation of Structure-Activity Relationships in Real-World Drug Design Data Sets Using Explainable Artificial Intelligence.

In silico models based on Deep Neural Networks (DNNs) are promising for predicting activities and properties of new molecules. Unfortunately, their inherent black-box character hinders our understanding, as to which structural features are important for activity. However, this information is crucial for capturing the underlying structure-activity relationships (SARs) to guide further optimization. To address this interpretation gap, "Explainable Artificial Intelligence" (XAI) methods recently became popular. Herein, we apply and compare multiple XAI methods to projects of lead optimization data sets with well-established SARs and available X-ray crystal structures. As we can show, easily understandable and comprehensive interpretations are obtained by combining DNN models with some powerful interpretation methods. In particular, SHAP-based methods are promising for this task. A novel visualization scheme using atom-based heatmaps provides useful insights into the underlying SAR. It is important to note that all interpretations are only meaningful in the context of the underlying models and associated data.

Quantum-mechanical property prediction of solvated drug molecules: what have we learned from a decade of SAMPL blind prediction challenges?

Joint academic-industrial projects supporting drug discovery are frequently pursued to deploy and benchmark cutting-edge methodical developments from academia in a real-world industrial environment at different scales. The dimensionality of tasks ranges from small molecule physicochemical property assessment over protein-ligand interaction up to statistical analyses of biological data. This way, method development and usability both benefit from insights gained at both ends, when predictiveness and readiness of novel approaches are confirmed, but the pharmaceutical drug makers get early access to novel tools for the quality of drug products and benefit of patients. Quantum-mechanical and simulation methods particularly fall into this group of methods, as they require skills and expense in their development but also significant resources in their application, thus are comparatively slowly dripping into the realm of industrial use. Nevertheless, these physics-based methods are becoming more and more useful. Starting with a general overview of these and in particular quantum-mechanical methods for drug discovery we review a decade-long and ongoing collaboration between Sanofi and the Kast group focused on the application of the embedded cluster reference interaction site model (EC-RISM), a solvation model for quantum chemistry, to study small molecule chemistry in the context of joint participation in several SAMPL (Statistical Assessment of Modeling of Proteins and Ligands) blind prediction challenges. Starting with early application to tautomer equilibria in water (SAMPL2) the methodology was further developed to allow for challenge contributions related to predictions of distribution coefficients (SAMPL5) and acidity constants (SAMPL6) over the years. Particular emphasis is put on a frequently overlooked aspect of measuring the quality of models, namely the retrospective analysis of earlier datasets and predictions in light of more recent and advanced developments. We therefore demonstrate the performance of the current methodical state of the art as developed and optimized for the SAMPL6 pKa and octanol-water log P challenges when re-applied to the earlier SAMPL5 cyclohexane-water log D and SAMPL2 tautomer equilibria datasets. Systematic improvement is not consistently found throughout despite the similarity of the problem class, i.e. protonation reactions and phase distribution. Hence, it is possible to learn about hidden bias in model assessment, as results derived from more elaborate methods do not necessarily improve quantitative agreement. This indicates the role of chance or coincidence for model development on the one hand which allows for the identification of systematic error and opportunities toward improvement and reveals possible sources of experimental uncertainty on the other. These insights are particularly useful for further academia-industry collaborations, as both partners are then enabled to optimize both the computational and experimental settings for data generation.

Using Graph Databases to Investigate Trends in Structure-Activity Relationship Networks.

Mining the steadily increasing amount of chemical and biological data is a key challenge in drug discovery. Graph databases offer viable alternatives for capturing interrelationships between molecules and for generating novel insights for design. In a graph database, molecules and their properties are mapped to nodes, while relationships are described by edges. Here, we introduce a graph database for navigation in chemical space, analogue searching, and structure-activity relationship (SAR) analysis. We illustrate this concept using hERG channel inhibitors from ChEMBL to extract SAR knowledge. This graph database is built using different relationships, namely 2D-fingerprint similarity, matched molecular pairs, topomer distances, and structure-activity landscape indices (SALI). Typical applications include retrieving analogues linked by single or multiple edge paths to the query compound as well as detection of nonadditive SAR features. Finally, we identify triplets of linked molecules for clustering. The speed of searching and analysis allows the user to interactively navigate the database and to address complex questions in real-time.

Impact of Different Automated Binding Pose Generation Approaches on Relative Binding Free Energy Simulations.

Relative binding free energy (RBFE) prediction methods such as free energy perturbation (FEP) are important today for estimating protein-ligand binding affinities. Significant hardware and algorithmic improvements now allow for simulating congeneric series within days. Therefore, RBFE calculations have an enormous potential for structure-based drug discovery. As typically only a few representative crystal structures for a series are available, other ligands and design proposals must be reliably superimposed for meaningful results. An observed significant effect of the alignment on FEP led us to develop an alignment approach combining docking with maximum common substructure (MCS) derived core constraints from the most similar reference pose, named MCS-docking workflow. We then studied the effect of binding pose generation on the accuracy of RBFE predictions using six ligand series from five pharmaceutically relevant protein targets. Overall, the MCS-docking workflow generated consistent poses for most of the ligands in the investigated series. While multiple alignment methods often resulted in comparable FEP predictions, for most of the cases herein, the MCS-docking workflow produced the best accuracy in predictions. Furthermore, the FEP analysis data strongly support the hypothesis that the accuracy of RBFE predictions depends on input poses to construct the perturbation map. Therefore, an automated workflow without manual intervention minimizes potential errors and obtains the most useful predictions with significant impact for structure-based design.

Artificial Intelligence in Drug Design.

Artificial Intelligence (AI) plays a pivotal role in drug discovery. In particular artificial neural networks such as deep neural networks or recurrent networks drive this area. Numerous applications in property or activity predictions like physicochemical and ADMET properties have recently appeared and underpin the strength of this technology in quantitative structure-property relationships (QSPR) or quantitative structure-activity relationships (QSAR). Artificial intelligence in de novo design drives the generation of meaningful new biologically active molecules towards desired properties. Several examples establish the strength of artificial intelligence in this field. Combination with synthesis planning and ease of synthesis is feasible and more and more automated drug discovery by computers is expected in the near future.

Thermodynamic Characterization of Hydration Sites from Integral Equation-Derived Free Energy Densities: Application to Protein Binding Sites and Ligand Series.

Water molecules play an essential role for mediating interactions between ligands and protein binding sites. Displacement of specific water molecules can favorably modulate the free energy of binding of protein-ligand complexes. Here, the nature of water interactions in protein binding sites is investigated by 3D RISM (three-dimensional reference interaction site model) integral equation theory to understand and exploit local thermodynamic features of water molecules by ranking their possible displacement in structure-based design. Unlike molecular dynamics-based approaches, 3D RISM theory allows for fast and noise-free calculations using the same detailed level of solute-solvent interaction description. Here we correlate molecular water entities instead of mere site density maxima with local contributions to the solvation free energy using novel algorithms. Distinct water molecules and hydration sites are investigated in multiple protein-ligand X-ray structures, namely streptavidin, factor Xa, and factor VIIa, based on 3D RISM-derived free energy density fields. Our approach allows the semiquantitative assessment of whether a given structural water molecule can potentially be targeted for replacement in structure-based design. Finally, PLS-based regression models from free energy density fields used within a 3D-QSAR approach (CARMa - comparative analysis of 3D RISM Maps) are shown to be able to extract relevant information for the interpretation of structure-activity relationship (SAR) trends, as demonstrated for a series of serine protease inhibitors.

The hpCADD NDDO Hamiltonian: Parametrization.

A neglect of diatomic differential overlap (NDDO) Hamiltonian has been parametrized as an electronic component of a polarizable force field. Coulomb and exchange potentials derived directly from the NDDO Hamiltonian in principle can be used with classical potentials, thus forming the basis for a new generation of efficiently applicable multipolar polarizable force fields. The new hpCADD Hamiltonian uses force-field-like atom types and reproduces the electrostatic properties (dipole moment, molecular electrostatic potential) and Koopmans' theorem ionization potentials closely, as demonstrated for a large training set and an independent test set of small molecules. The Hamiltonian is not intended to reproduce geometries or total energies well, as these will be controlled by the classical force-field potentials. In order to establish the hpCADD Hamiltonian as an electronic component in force-field-based calculations, we tested its performance in combination with the 3D reference interaction site model (3D RISM) for aqueous solutions. Comparison of the resulting solvation free energies for the training and test sets to atomic charges derived from standard procedures, exact solute-solvent electrostatics based on high-level quantum-chemical reference data, and established semiempirical Hamiltonians demonstrates the advantages of the hpCADD parametrization.

A combination of spin diffusion methods for the determination of protein-ligand complex structural ensembles.

Structure-based drug design (SBDD) is a powerful and widely used approach to optimize affinity of drug candidates. With the recently introduced INPHARMA method, the binding mode of small molecules to their protein target can be characterized even if no spectroscopic information about the protein is known. Here, we show that the combination of the spin-diffusion-based NMR methods INPHARMA, trNOE, and STD results in an accurate scoring function for docking modes and therefore determination of protein-ligand complex structures. Applications are shown on the model system protein kinase A and the drug targets glycogen phosphorylase and soluble epoxide hydrolase (sEH). Multiplexing of several ligands improves the reliability of the scoring function further. The new score allows in the case of sEH detecting two binding modes of the ligand in its binding site, which was corroborated by X-ray analysis.

Targeting dynamic pockets of HIV-1 protease by structure-based computational screening for allosteric inhibitors.

We present the discovery of low molecular weight inhibitors of human immunodeficiency virus 1 (HIV-1) protease subtype B that were identified by structure-based virtual screening as ligands of an allosteric surface cavity. For pocket identification and prioritization, we performed a molecular dynamics simulation and observed several flexible, partially transient surface cavities. For one of these presumable ligand-binding pockets that are located in the so-called "hinge region" of the identical protease chains, we computed a receptor-derived pharmacophore model, with which we retrieved fragment-like inhibitors from a screening compound pool. The most potent hit inhibited protease activity in vitro in a noncompetitive mode of action. Although attempts failed to crystallize this ligand bound to the enzyme, the study provides proof-of-concept for identifying innovative tool compounds for chemical biology by addressing flexible protein models with receptor pocket-derived pharmacophore screening.

Back in Person: Frontiers in Medicinal Chemistry 2023.

The Frontiers in Medicinal Chemistry (FiMC) is the largest international Medicinal Chemistry conference in the German speaking area and took place from April 3rd to 5th 2023 in Vienna (Austria). Fortunately, after being cancelled in 2020 and two years (2021-2022) of entirely virtual meetings, due to the COVID-19 pandemic, the FiMC could be held in a face-to-face format again. Organized by the Division of Medicinal Chemistry of the German Chemical Society (GDCh), the Division of Pharmaceutical and Medicinal Chemistry of the German Pharmaceutical Society (DPhG), together with the Division of Medicinal Chemistry of the Austrian Chemical Society (GÖCH), the Austrian Pharmaceutical Society (ÖPhG), and a local organization committee from the University of Vienna headed by Thierry Langer, the meeting brought together 260 participants from 21 countries. The program included 38 lectures by leading scientists from industry and academia as well as early career investigators. Moreover, 102 posters were presented in two highly interactive poster sessions.

3D-QSAR based on quantum-chemical molecular fields: toward an improved description of halogen interactions.

Current 3D-QSAR methods such as CoMFA or CoMSIA make use of classical force-field approaches for calculating molecular fields. Thus, they can not adequately account for noncovalent interactions involving halogen atoms like halogen bonds or halogen-π interactions. These deficiencies in the underlying force fields result from the lack of treatment of the anisotropy of the electron density distribution of those atoms, known as the "σ-hole", although recent developments have begun to take specific interactions such as halogen bonding into account. We have now replaced classical force field derived molecular fields by local properties such as the local ionization energy, local electron affinity, or local polarizability, calculated using quantum-mechanical (QM) techniques that do not suffer from the above limitation for 3D-QSAR. We first investigate the characteristics of QM-based local property fields to show that they are suitable for statistical analyses after suitable pretreatment. We then analyze these property fields with partial least-squares (PLS) regression to predict biological affinities of two data sets comprising factor Xa and GABA-A/benzodiazepine receptor ligands. While the resulting models perform equally well or even slightly better in terms of consistency and predictivity than the classical CoMFA fields, the most important aspect of these augmented field-types is that the chemical interpretation of resulting QM-based property field models reveals unique SAR trends driven by electrostatic and polarizability effects, which cannot be extracted directly from CoMFA electrostatic maps. Within the factor Xa set, the interaction of chlorine and bromine atoms with a tyrosine side chain in the protease S1 pocket are correctly predicted. Within the GABA-A/benzodiazepine ligand data set, PLS models of high predictivity resulted for our QM-based property fields, providing novel insights into key features of the SAR for two receptor subtypes and cross-receptor selectivity of the ligands. The detailed interpretation of regression models derived using improved QM-derived property fields thus provides a significant advantage by revealing chemically meaningful correlations with biological activity and helps in understanding novel structure-activity relationship features. This will allow such knowledge to be used to design novel molecules on the basis of interactions additional to steric and hydrogen-bonding features.

Development of in silico filters to predict activation of the pregnane X receptor (PXR) by structurally diverse drug-like molecules.

The pregnane X receptor (PXR), a member of the nuclear hormone superfamily, regulates the expression of several enzymes and transporters involved in metabolically relevant processes. The significant induction of CYP450 enzymes by PXR, in particular CYP3A4, might significantly alter the metabolism of prescribed drugs. In order to early identify molecules in drug discovery with a potential to activate PXR as antitarget, we developed fast and reliable in silico filters by ligand-based QSAR techniques. Two classification models were established on a diverse dataset of 434 drug-like molecules. A second augmented set allowed focusing on interesting regions in chemical space. These classifiers are based on decision trees combined with a genetic algorithm based variable selection to arrive at predictive models. The classifier for the first dataset on 29 descriptors showed good performance on a test set with a correct classification of both 100% for PXR activators and non-activators plus 87% for activators and 83% for non-activators in an external dataset. The second classifier then correctly predicts 97% activators and 91% non-activators in a test set and 94% for activators and 64% non-activators in an external set of 50 molecules, which still qualifies for application as a filter focusing on PXR activators. Finally a quantitative model for PXR activation for a subset of these molecules was derived using a regression-tree approach combined with GA variable selection. This final model shows a predictive r(2) of 0.774 for the test set and 0.452 for an external set of 33 molecules. Thus, the combination of these filters consistently provide guidelines for lowering PXR activation in novel candidate molecules.

Optimizing interactions to protein binding sites by integrating docking-scoring strategies into generative AI methods.

The identification and optimization of promising lead molecules is essential for drug discovery. Recently, artificial intelligence (AI) based generative methods provided complementary approaches for generating molecules under specific design constraints of relevance in drug design. The goal of our study is to incorporate protein 3D information directly into generative design by flexible docking plus an adapted protein-ligand scoring function, thereby moving towards automated structure-based design. First, the protein-ligand scoring function RFXscore integrating individual scoring terms, ligand descriptors, and combined terms was derived using the PDBbind database and internal data. Next, design results for different workflows are compared to solely ligand-based reward schemes. Our newly proposed, optimal workflow for structure-based generative design is shown to produce promising results, especially for those exploration scenarios, where diverse structures fitting to a protein binding site are requested. Best results are obtained using docking followed by RFXscore, while, depending on the exact application scenario, it was also found useful to combine this approach with other metrics that bias structure generation into "drug-like" chemical space, such as target-activity machine learning models, respectively.

Artificial Intelligence in Compound Design.

Artificial intelligence has seen an incredibly fast development in recent years. Many novel technologies for property prediction of drug molecules as well as for the design of novel molecules were introduced by different research groups. These artificial intelligence-based design methods can be applied for suggesting novel chemical motifs in lead generation or scaffold hopping as well as for optimization of desired property profiles during lead optimization. In lead generation, broad sampling of the chemical space for identification of novel motifs is required, while in the lead optimization phase, a detailed exploration of the chemical neighborhood of a current lead series is advantageous. These different requirements for successful design outcomes render different combinations of artificial intelligence technologies useful. Overall, we observe that a combination of different approaches with tailored scoring and evaluation schemes appears beneficial for efficient artificial intelligence-based compound design.

CACHE (Critical Assessment of Computational Hit-finding Experiments): A public-private partnership benchmarking initiative to enable the development of computational methods for hit-finding.

One aspirational goal of computational chemistry is to predict potent and drug-like binders for any protein, such that only those that bind are synthesized. In this Roadmap, we describe the launch of Critical Assessment of Computational Hit-finding Experiments (CACHE), a public benchmarking project to compare and improve small molecule hit-finding algorithms through cycles of prediction and experimental testing. Participants will predict small molecule binders for new and biologically relevant protein targets representing different prediction scenarios. Predicted compounds will be tested rigorously in an experimental hub, and all predicted binders as well as all experimental screening data, including the chemical structures of experimentally tested compounds, will be made publicly available, and not subject to any intellectual property restrictions. The ability of a range of computational approaches to find novel binders will be evaluated, compared, and openly published. CACHE will launch 3 new benchmarking exercises every year. The outcomes will be better prediction methods, new small molecule binders for target proteins of importance for fundamental biology or drug discovery, and a major technological step towards achieving the goal of Target 2035, a global initiative to identify pharmacological probes for all human proteins.

Application of Deep Neural Network Models in Drug Discovery Programs.

In silico driven optimization of compound properties related to pharmacokinetics, pharmacodynamics, and safety is a key requirement in modern drug discovery. Nowadays, large and harmonized datasets allow to implement deep neural networks (DNNs) as a framework for leveraging predictive models. Nevertheless, various available model architectures differ in their global applicability and performance in lead optimization projects, such as stability over time and interpretability of the results. Here, we describe and compare the value of established DNN-based methods for the prediction of key ADME property trends and biological activity in an industrial drug discovery environment, represented by microsomal lability, CYP3A4 inhibition and factor Xa inhibition. Three architectures are exemplified, our earlier described multilayer perceptron approach (MLP), graph convolutional network-based models (GCN) and a vector representation approach, Mol2Vec. From a statistical perspective, MLP and GCN were found to perform superior over Mol2Vec, when applied to external validation sets. Interestingly, GCN-based predictions are most stable over a longer period in a time series validation study. Apart from those statistical observations, DNN prove of value to guide local SAR. To illustrate this important aspect in pharmaceutical research projects, we discuss challenging applications in medicinal chemistry towards a more realistic picture of artificial intelligence in drug discovery.

A Solid Supported Membrane-Based Technology for Electrophysical Screening of B0AT1-Modulating Compounds.

Classical high-throughput screening (HTS) technologies for the analysis of ionic currents across biological membranes can be performed using fluorescence-based, radioactive, and mass spectrometry (MS)-based uptake assays. These assays provide rapid results for pharmacological HTS, but the underlying, indirect analytical character of these assays can be linked to high false-positive hit rates. Thus, orthogonal and secondary assays using more biological target-based technologies are indispensable for further compound validation and optimization. Direct assay technologies for transporter proteins are electrophysiology-based, but are also complex, time-consuming, and not well applicable for automated profiling purposes. In contrast to conventional patch clamp systems, solid supported membrane (SSM)-based electrophysiology is a sensitive, membrane-based method for transporter analysis, and current technical developments target the demand for automated, accelerated, and sensitive assays for transporter-directed compound screening. In this study, the suitability of the SSM-based technique for pharmacological compound identification and optimization was evaluated performing cell-free SSM-based measurements with the electrogenic amino acid transporter B0AT1 (SLC6A19). Electrophysiological characterization of leucine-induced currents demonstrated that the observed signals were specific to B0AT1. Moreover, B0AT1-dependent responses were successfully inhibited using an established in-house tool compound. Evaluation of current stability and data reproducibility verified the robustness and reliability of the applied assay. Active compounds from primary screens of large compound libraries were validated, and false-positive hits were identified. These results clearly demonstrate the suitability of the SSM-based technique as a direct electrophysiological method for rapid and automated identification of small molecules that can inhibit B0AT1 activity.

The PROTACtable genome.

Proteolysis-targeting chimeras (PROTACs) are an emerging drug modality that may offer new opportunities to circumvent some of the limitations associated with traditional small-molecule therapeutics. By analogy with the concept of the 'druggable genome', the question arises as to which potential drug targets might PROTAC-mediated protein degradation be most applicable. Here, we present a systematic approach to the assessment of the PROTAC tractability (PROTACtability) of protein targets using a series of criteria based on data and information from a diverse range of relevant publicly available resources. Our approach could support decision-making on whether or not a particular target may be amenable to modulation using a PROTAC. Using our approach, we identified 1,067 proteins of the human proteome that have not yet been described in the literature as PROTAC targets that offer potential opportunities for future PROTAC-based efforts.