Druggability Assessment in TRAPP Using Machine Learning Approaches

Toward the Prediction of Binding Events in Very Flexible, Allosteric, Multidomain Proteins

Knowledge of the structures formed by proteins and small molecules is key to understand the molecular principles of chemotherapy and for designing new and more effective drugs. During the early stage of a drug discovery program, it is customary to predict ligand-protein complexes in silico, particularly when screening large compound databases. While virtual screening based on molecular docking is widely used for this purpose, it generally fails in mimicking binding events associated with large conformational changes in the protein, particularly when the latter involve multiple domains. In this work, we describe a new methodology to generate bound-like conformations of very flexible and allosteric proteins bearing multiple binding sites by exploiting only information on the unbound structure and the putative binding sites. The protocol is validated on the paradigm enzyme adenylate kinase, for which we generated a significant fraction of bound-like structures. A fraction of these conformations, employed in ensemble-docking calculations, allowed to find native-like poses of substrates and inhibitors (binding to the active form of the enzyme), as well as catalytically incompetent analogs (binding the inactive form). Our protocol provides a general framework for the generation of bound-like conformations of challenging drug targets that are suitable to host different ligands, demonstrating high sensitivity to the fine chemical details that regulate protein's activity. We foresee applications in virtual screening, in the prediction of the impact of amino acid mutations on structure and dynamics, and in protein engineering.

dyphAI dynamic pharmacophore modeling with AI: a tool for efficient screening of new acetylcholinesterase inhibitors

Therapeutic strategies for Alzheimer's disease (AD) often involve inhibiting acetylcholinesterase (AChE), underscoring the need for novel inhibitors with high selectivity and minimal side effects. A detailed analysis of the protein-ligand pharmacophore dynamics can facilitate this. In this study, we developed and employed dyphAI, an innovative approach integrating machine learning models, ligand-based pharmacophore models, and complex-based pharmacophore models into a pharmacophore model ensemble. This ensemble captures key protein-ligand interactions, including π-cation interactions with Trp-86 and several π-π interactions with residues Tyr-341, Tyr-337, Tyr-124, and Tyr-72. The protocol identified 18 novel molecules from the ZINC database with binding energy values ranging from -62 to -115 kJ/mol, suggesting their strong potential as AChE inhibitors. To further validate the predictions, nine molecules were acquired and tested for their inhibitory activity against human AChE. Experimental results revealed that molecules, 4 (P-1894047), with its complex multi-ring structure and numerous hydrogen bond acceptors, and 7 (P-2652815), characterized by a flexible, polar framework with ten hydrogen bond donors and acceptors, exhibited IC₅₀ values lower than or equal to that of the control (galantamine), indicating potent inhibitory activity. Similarly, molecules 5 (P-1205609), 6 (P-1206762), 8 (P-2026435), and 9 (P-533735) also demonstrated strong inhibition. In contrast, molecule 3 (P-617769798) showed a higher IC50 value, and molecules 1 (P-14421887) and 2 (P-25746649) yielded inconsistent results, likely due to solubility issues in the experimental setup. These findings underscore the value of integrating computational predictions with experimental validation, enhancing the reliability of virtual screening in the discovery of potent enzyme inhibitors.

Mapping the Binding Hotspots and Transient Binding Pockets on V-Domain Immunoglobulin Suppressor of T Cell Activation Protein Surface

V-domain immunoglobulin suppressor of T cell activation (VISTA), an inhibitory immune checkpoint present on both immune and tumor cells, has emerged as a highly promising target for cancer therapy due to its potential to overcome resistance encountered with existing immune checkpoint treatments. VSIG-3 is determined as an inhibitory ligand for VISTA, leading to the suppression of T cell proliferation. However, hotspots between VISTA/VSIG-3 protein-protein interaction remain ambiguous, mainly attributed to the lack of the structure of the VISTA/VSIG-3 complex. Therefore, in this study, in order to determine the energetic contributions of the interfacial residues on VISTA, we first constructed VISTA/VSIG-3 complex models by the protein docking method, followed by molecular dynamics simulations, binding free-energy decomposition, and alanine scanning. Results suggested that the putative hotspots in VISTA comprise residues His32, Tyr37, Thr35, Glu47, Val48, Gln49, Glu53, Arg54, Gln73, His122, and His126. Moreover, the distribution of the hotspots was clustered into two regions (hot regions I and II), and by using the TRAPP tool, transient subpockets within the hot regions were identified. Furthermore, conformational states of the binding pockets exhibiting druggability scores higher than those observed in the crystal structure were found. Overall, we hope that the findings outlined in this study can be used to facilitate the development of inhibitors targeting the VISTA/VSIG-3 immune checkpoint pathway in the future.

Repurposing FDA-Approved Drugs Against Potential Drug Targets Involved in Brain Inflammation Contributing to Alzheimer's Disease

Alzheimer's disease is a neurodegenerative disease that continues to have a rising number of cases. While extensive research has been conducted in the last few decades, only a few drugs have been approved by the FDA for treatment, and even fewer aim to be curative rather than manage symptoms. There remains an urgent need for understanding disease pathogenesis, as well as identifying new targets for further drug discovery. Alzheimer's disease (AD) is known to stem from a build-up of amyloid beta (Aβ) plaques as well as tangles of tau proteins. Furthermore, inflammation in the brain is known to arise from the degeneration of tissue and the build-up of insoluble material. Therefore, there is a potential link between the pathology of AD and inflammation in the brain, especially as the disease progresses to later stages where neuronal death and degeneration levels are higher. Proteins that are relevant to both brain inflammation and AD thus make ideal potential targets for therapeutics; however, the proteins need to be evaluated to determine which targets would be ideal for potential drug therapeutic treatments, or 'druggable'. Druggability analysis was conducted using two structure-based methods (i.e., Drug-Like Density analysis and SiteMap), as well as a sequence-based approach, SPIDER. The most druggable targets were then evaluated using single-nuclei sequencing data for their clinical relevance to inflammation in AD. For each of the top five targets, small molecule docking was used to evaluate which FDA approved drugs were able to bind with the chosen proteins. The top targets included DRD2 (inhibits adenylyl cyclase activity), C9 (binds with C5B8 to form the membrane attack complex), C4b (binds with C2a to form C3 convertase), C5AR1 (GPCR that binds C5a), and GABA-A-R (GPCR involved in inhibiting neurotransmission). Each target had multiple potential inhibitors from the FDA-approved drug list with decent binding infinities. Among these inhibitors, two drugs were found as top inhibitors for more than one protein target. They are C15H14N2O2 and v316 (Paracetamol), used to treat pain/inflammation originally for cataracts and relieve headaches/fever, respectively. These results provide the groundwork for further experimental investigation or clinical trials.

Predicting binding events in very flexible, allosteric, multi-domain proteins

Knowledge of the structures formed by proteins and small molecules is key to understand the molecular principles of chemotherapy and for designing new and more effective drugs. During the early stage of a drug discovery program, it is customary to predict ligand-protein complexes in silico, particularly when screening large compound databases. While virtual screening based on molecular docking is widely used for this purpose, it generally fails in mimicking binding events associated with large conformational changes in the protein, particularly when the latter involve multiple domains. In this work, we describe a new methodology to generate bound-like conformations of very flexible and allosteric proteins bearing multiple binding sites by exploiting only information on the unbound structure and the putative binding sites. The protocol is validated on the paradigm enzyme adenylate kinase, for which we generated a significant fraction of bound-like structures. A fraction of these conformations, employed in ensemble-docking calculations, allowed to find native-like poses of substrates and inhibitors (binding to the active form of the enzyme), as well as catalytically incompetent analogs (binding the inactive form). Our protocol provides a general framework for the generation of bound-like conformations of challenging drug targets that are suitable to host different ligands, demonstrating high sensitivity to the fine chemical details that regulate protein's activity. We foresee applications in virtual screening, in the prediction of the impact of amino acid mutations on structure and dynamics, and in protein engineering.

Automated Flow Peptide Synthesis Enables Engineering of Proteins with Stabilized Transient Binding Pockets

Engineering at the amino acid level is key to enhancing the properties of existing proteins in a desired manner. So far, protein engineering has been dominated by genetic approaches, which have been extremely powerful but only allow for minimal variations beyond the canonical amino acids. Chemical peptide synthesis allows the unrestricted incorporation of a vast set of unnatural amino acids with much broader functionalities, including the incorporation of post-translational modifications or labels. Here we demonstrate the potential of chemical synthesis to generate proteins in a specific conformation, which would have been unattainable by recombinant protein expression. We use recently established rapid automated flow peptide synthesis combined with solid-phase late-stage modifications to rapidly generate a set of FK506-binding protein 51 constructs bearing defined intramolecular lactam bridges. This trapped an otherwise rarely populated transient pocket-as confirmed by crystal structures-which led to an up to 39-fold improved binding affinity for conformation-selective ligands and represents a unique system for the development of ligands for this rare conformation. Overall, our results show how rapid automated flow peptide synthesis can be applied to precision protein engineering.

Molecular Dynamics Simulations of the Human Ecto-5'-Nucleotidase (h-ecto-5'-NT, CD73): Insights into Protein Flexibility and Binding Site Dynamics

Human ecto-5'-nucleotidase (h-ecto-5'-NT, CD73) is a homodimeric Zn2+-binding metallophosphoesterase that hydrolyzes adenosine 5'-monophosphate (5'-AMP) to adenosine and phosphate. h-Ecto-5'-NT is a key enzyme in purinergic signaling pathways and has been recognized as a promising biological target for several diseases, including cancer and inflammatory, infectious, and autoimmune diseases. Despite its importance as a biological target, little is known about h-ecto-5'-NT dynamics, which poses a considerable challenge to the design of inhibitors of this target enzyme. Here, to explore h-ecto-5'-NT flexibility, all-atom unbiased molecular dynamics (MD) simulations were performed. Remarkable differences in the dynamics of the open (catalytically inactive) and closed (catalytically active) conformations of the apo-h-ecto-5'-NT were observed during the simulations, and the nucleotide analogue inhibitor AMPCP was shown to stabilize the protein structure in the closed conformation. Our results suggest that the large and complex domain motion that enables the h-ecto-5'-NT open/closed conformational switch is slow, and therefore, it could not be completely captured within the time scale of our simulations. Nonetheless, we were able to explore the faster dynamics of the h-ecto-5'-NT substrate binding site, which is mainly located at the C-terminal domain and well conserved among the protein's open and closed conformations. Using the TRAPP ("Transient Pockets in Proteins") approach, we identified transient subpockets close to the substrate binding site. Finally, conformational states of the substrate binding site with higher druggability scores than the crystal structure were identified. In summary, our study provides valuable insights into h-ecto-5'-NT structural flexibility, which can guide the structure-based design of novel h-ecto-5'-NT inhibitors.

DrugnomeAI is an ensemble machine-learning framework for predicting druggability of candidate drug targets

The druggability of targets is a crucial consideration in drug target selection. Here, we adopt a stochastic semi-supervised ML framework to develop DrugnomeAI, which estimates the druggability likelihood for every protein-coding gene in the human exome. DrugnomeAI integrates gene-level properties from 15 sources resulting in 324 features. The tool generates exome-wide predictions based on labelled sets of known drug targets (median AUC: 0.97), highlighting features from protein-protein interaction networks as top predictors. DrugnomeAI provides generic as well as specialised models stratified by disease type or drug therapeutic modality. The top-ranking DrugnomeAI genes were significantly enriched for genes previously selected for clinical development programs (p value < 1 × 10-308) and for genes achieving genome-wide significance in phenome-wide association studies of 450 K UK Biobank exomes for binary (p value = 1.7 × 10-5) and quantitative traits (p value = 1.6 × 10-7). We accompany our method with a web application ( http://drugnomeai.public.cgr.astrazeneca.com ) to visualise the druggability predictions and the key features that define gene druggability, per disease type and modality.

The druggable genome: Twenty years later

The concept of the druggable genome has been with us for 20 years. During this time, researchers have developed several methods and resources to help assess a target's druggability. In parallel, evidence for target-disease associations has been collated at scale by Open Targets. More recently, the Protein Data Bank in Europe (PDBe) have built a knowledge base matching per-residue annotations with available protein structure. While each resource is useful in isolation, we believe there is enormous potential in bringing all relevant data into a single knowledge graph, from gene-level to protein residue. Automation is vital for the processing and assessment of all available structures. We have developed scalable, automated workflows that provide hotspot-based druggability assessments for all available structures across large numbers of targets. Ultimately, we will run our method at a proteome scale, an ambition made more realistic by the arrival of AlphaFold 2. Bringing together annotations from the residue up to the gene level and building connections within the graph to represent pathways or protein-protein interactions will create complexity that mirrors the biological systems they represent. Such complexity is difficult for the human mind to utilise effectively, particularly at scale. We believe that graph-based AI methods will be able to expertly navigate such a knowledge graph, selecting the targets of the future.

Discovery of the Cryptic Sites of SARS-CoV-2 Papain-like Protease and Analysis of Its Druggability

In late 2019, a new coronavirus (CoV) caused the outbreak of a deadly respiratory disease, resulting in the COVID-19 pandemic. In view of the ongoing pandemic, there is an immediate need to find drugs to treat patients. SARS-CoV-2 papain-like cysteine protease (PLpro) not only plays an important role in the pathogenesis of the virus but is also a target protein for the development of inhibitor drugs. Therefore, to develop targeted inhibitors, it is necessary to analyse and verify PLpro sites and explore whether there are other cryptic binding pockets with better activity. In this study, first, we detected the site of the whole PLpro protein by sitemap of Schrödinger (version 2018), the cavity of LigBuilder V3, and DeepSite, and roughly judged the possible activated binding site area. Then, we used the mixed solvent dynamics simulation (MixMD) of probe molecules to induce conformational changes in the protein to find the possible cryptic active sites. Finally, the TRAPP method was used to predict the druggability of cryptic pockets and analyse the changes in the physicochemical properties of residues around these sites. This work will help promote the research of SARS-CoV-2 PLpro inhibitors.

Probabilistic Pocket Druggability Prediction via One-Class Learning

The choice of target pocket is a key step in a drug discovery campaign. This step can be supported by in silico druggability prediction. In the literature, druggability prediction is often approached as a two-class classification task that distinguishes between druggable and non-druggable (or less druggable) pockets (or voxels). Apart from obvious cases, however, the non-druggable class is conceptually ambiguous. This is because any pocket (or target) is only non-druggable until a drug is found for it. It is therefore more appropriate to adopt a one-class approach, which uses only unambiguous information, namely, druggable pockets. Here, we propose using the import vector domain description (IVDD) algorithm to support this task. IVDD is a one-class probabilistic kernel machine that we previously introduced. To feed the algorithm, we use customized DrugPred descriptors computed via NanoShaper. Our results demonstrate the feasibility and effectiveness of the approach. In particular, we can remove or mitigate biases chiefly due to the labeling.

Artificial intelligence and machine-learning approaches in structure and ligand-based discovery of drugs affecting central nervous system

CNS disorders are indications with a very high unmet medical needs, relatively smaller number of available drugs, and a subpar satisfaction level among patients and caregiver. Discovery of CNS drugs is extremely expensive affair with its own unique challenges leading to extremely high attrition rates and low efficiency. With explosion of data in information age, there is hardly any aspect of life that has not been touched by data driven technologies such as artificial intelligence (AI) and machine learning (ML). Drug discovery is no exception, emergence of big data via genomic, proteomic, biological, and chemical technologies has driven pharmaceutical giants to collaborate with AI oriented companies to revolutionise drug discovery, with the goal of increasing the efficiency of the process. In recent years many examples of innovative applications of AI and ML techniques in CNS drug discovery has been reported. Research on therapeutics for diseases such as schizophrenia, Alzheimer's and Parkinsonism has been provided with a new direction and thrust from these developments. AI and ML has been applied to both ligand-based and structure-based drug discovery and design of CNS therapeutics. In this review, we have summarised the general aspects of AI and ML from the perspective of drug discovery followed by a comprehensive coverage of the recent developments in the applications of AI/ML techniques in CNS drug discovery.

P2X3-selective mechanism of Gefapixant, a drug candidate for the treatment of refractory chronic cough

•

The mechanism by which Gefapixant/AF-219 selectively acts on the P2X3 receptor is unclear.

•

The negative allosteric site of AF-219 at P2X3 is also a potent allosteric site for other P2X subtypes.

•

The selectivity of AF-219 for P2X3 is determined by the accessibility of binding site and the internal shape of this pocket.

•

The finding will provide new perspectives for drug design against P2X3-mediated diseases such as RCC.

Gefapixant/AF-219, a selective inhibitor of the P2X3 receptor, is the first new drug other than dextromethorphan to be approved for the treatment of refractory chronic cough (RCC) in nearly 60 years. To date, seven P2X subtypes (P2X1-7) activated by extracellular ATP have been cloned, and subtype selectivity of P2X inhibitors is a prerequisite for reducing side effects. We previously identified the site and mechanism of action of Gefapixant/AF-219 on the P2X3 receptor, which occupies a pocket consisting of the left flipper (LF) and lower body (LB) domains. However, the mechanism by which AF-219 selectively acts on the P2X3 receptor is unknown. Here, we combined mutagenesis, chimera construction, molecular simulations, covalent occupation and chemical synthesis, and find that the negative allosteric site of AF-219 at P2X3 is also present in other P2X subtypes, at least for P2X1, P2X2 and P2X4. By constructing each chimera of AF-219 sensitive P2X3 and insensitive P2X2 subtypes, the insensitive P2X2 subtype was made to acquire the inhibitory properties of AF-219 and AF-353, an analog of AF-219 with higher affinity. Our results suggest that the selectivity of AF-219/AF-353 for P2X3 over the other P2X subtypes is determined by a combination of the accessibility of P2X3 binding site and the internal shape of this pocket, a finding that could provide new perspectives for drug design against P2X3-mediated diseases such as RCC, idiopathic pulmonary fibrosis, hypertension and overactive bladder disorder.

SCoV2-MD: a database for the dynamics of the SARS-CoV-2 proteome and variant impact predictions

SCoV2-MD (www.scov2-md.org) is a new online resource that systematically organizes atomistic simulations of the SARS-CoV-2 proteome. The database includes simulations produced by leading groups using molecular dynamics (MD) methods to investigate the structure-dynamics-function relationships of viral proteins. SCoV2-MD cross-references the molecular data with the pandemic evolution by tracking all available variants sequenced during the pandemic and deposited in the GISAID resource. SCoV2-MD enables the interactive analysis of the deposited trajectories through a web interface, which enables users to search by viral protein, isolate, phylogenetic attributes, or specific point mutation. Each mutation can then be analyzed interactively combining static (e.g. a variety of amino acid substitution penalties) and dynamic (time-dependent data derived from the dynamics of the local geometry) scores. Dynamic scores can be computed on the basis of nine non-covalent interaction types, including steric properties, solvent accessibility, hydrogen bonding, and other types of chemical interactions. Where available, experimental data such as antibody escape and change in binding affinities from deep mutational scanning experiments are also made available. All metrics can be combined to build predefined or custom scores to interrogate the impact of evolving variants on protein structure and function.

A Blueprint for High Affinity SARS-CoV-2 Mpro Inhibitors from Activity-Based Compound Library Screening Guided by Analysis of Protein Dynamics

The SARS-CoV-2 coronavirus outbreak continues to spread at a rapid rate worldwide. The main protease (Mpro) is an attractive target for anti-COVID-19 agents. Unexpected difficulties have been encountered in the design of specific inhibitors. Here, by analyzing an ensemble of ∼30 000 SARS-CoV-2 Mpro conformations from crystallographic studies and molecular simulations, we show that small structural variations in the binding site dramatically impact ligand binding properties. Hence, traditional druggability indices fail to adequately discriminate between highly and poorly druggable conformations of the binding site. By performing ∼200 virtual screenings of compound libraries on selected protein structures, we redefine the protein's druggability as the consensus chemical space arising from the multiple conformations of the binding site formed upon ligand binding. This procedure revealed a unique SARS-CoV-2 Mpro blueprint that led to a definition of a specific structure-based pharmacophore. The latter explains the poor transferability of potent SARS-CoV Mpro inhibitors to SARS-CoV-2 Mpro, despite the identical sequences of the active sites. Importantly, application of the pharmacophore predicted novel high affinity inhibitors of SARS-CoV-2 Mpro, that were validated by in vitro assays performed here and by a newly solved X-ray crystal structure. These results provide a strong basis for effective rational drug design campaigns against SARS-CoV-2 Mpro and a new computational approach to screen protein targets with malleable binding sites.

Med Chem Remote: The Frontiers in Medicinal Chemistry 2021

Digital, but delicious! The Frontiers in Medicinal Chemistry 2021 meeting, originally intended to take place in Darmstadt, carried on as an online event from March 8-10 this year. Even with pandemic restrictions, the event co-presented by the Medicinal Chemistry Division of the German Chemical Society (GDCh), the German Pharmaceutical Society (DPhG), and the Swiss Chemical Society (SCS) proved to be a success, showcasing excellent speakers and facilitating participant interaction in an ingenious virtual setting. Over 350 participants from more than 10 countries gathered to discuss the latest trends and directions in medicinal chemistry, with sessions on molecular glues, covalent fragments, transient binding pockets and more. This report presents a summary of the key lectures and activities at the event.

Finding Druggable Sites in Proteins Using TACTICS

Structure-based drug discovery efforts require knowledge of where drug-binding sites are located on target proteins. To address the challenge of finding druggable sites, we developed a machine-learning algorithm called TACTICS (trajectory-based analysis of conformations to identify cryptic sites), which uses an ensemble of molecular structures (such as molecular dynamics simulation data) as input. First, TACTICS uses k-means clustering to select a small number of conformations that represent the overall conformational heterogeneity of the data. Then, TACTICS uses a random forest model to identify potentially bindable residues in each selected conformation, based on protein motion and geometry. Lastly, residues in possible binding pockets are scored using fragment docking. As proof-of-principle, TACTICS was applied to the analysis of simulations of the SARS-CoV-2 main protease and methyltransferase and the Yersinia pestis aryl carrier protein. Our approach recapitulates known small-molecule binding sites and predicts the locations of sites not previously observed in experimentally determined structures. The TACTICS code is available at https://github.com/Albert-Lau-Lab/tactics_protein_analysis.

A 'deep dive' into the SARS-Cov-2 polymerase assembly: identifying novel allosteric sites and analyzing the hydrogen bond networks and correlated dynamics

Replication of the SARS-CoV-2 genome is a fundamental step in the virus life cycle and inhibiting the SARS-CoV2 replicase machinery has been proven recently as a promising approach in combating the virus. Despite this recent success, there are still several aspects related to the structure, function and dynamics of the CoV-2 polymerase that still need to be addressed. This includes understanding the dynamicity of the various polymerase subdomains, analyzing the hydrogen bond networks at the active site and at the template entry in the presence of water, studying the binding modes of the nucleotides at the active site, highlighting positions for acceptable nucleotides' substitutions that can be tolerated at different positions within the nascent RNA strand, identifying possible allosteric sites within the polymerase structure and studying their correlated dynamics relative to the catalytic site. Here, we combined various cutting-edge modelling tools with the recently resolved SARS-CoV-2 cryo-EM polymerase structures to fill this gap in knowledge. Our findings provide a detailed analysis of the hydrogen bond networks at various parts of the polymerase structure and suggest possible nucleotides' substitutions that can be tolerated by the polymerase complex. We also report here three 'druggable' allosteric sites within the NSP12 RdRp that can be targeted by small molecule inhibitors. Our correlated motion analysis shows that the dynamics within one of the newly identified sites are linked to the active site, indicating that targeting this site can significantly impact the catalytic activity of the SARS-CoV-2 polymerase.Communicated by Ramaswamy H. Sarma.

Artificial intelligence and machine learning-aided drug discovery in central nervous system diseases: State-of-the-arts and future directions

Neurological disorders significantly outnumber diseases in other therapeutic areas. However, developing drugs for central nervous system (CNS) disorders remains the most challenging area in drug discovery, accompanied with the long timelines and high attrition rates. With the rapid growth of biomedical data enabled by advanced experimental technologies, artificial intelligence (AI) and machine learning (ML) have emerged as an indispensable tool to draw meaningful insights and improve decision making in drug discovery. Thanks to the advancements in AI and ML algorithms, now the AI/ML-driven solutions have an unprecedented potential to accelerate the process of CNS drug discovery with better success rate. In this review, we comprehensively summarize AI/ML-powered pharmaceutical discovery efforts and their implementations in the CNS area. After introducing the AI/ML models as well as the conceptualization and data preparation, we outline the applications of AI/ML technologies to several key procedures in drug discovery, including target identification, compound screening, hit/lead generation and optimization, drug response and synergy prediction, de novo drug design, and drug repurposing. We review the current state-of-the-art of AI/ML-guided CNS drug discovery, focusing on blood-brain barrier permeability prediction and implementation into therapeutic discovery for neurological diseases. Finally, we discuss the major challenges and limitations of current approaches and possible future directions that may provide resolutions to these difficulties.

RASPD+: Fast Protein-Ligand Binding Free Energy Prediction Using Simplified Physicochemical Features

The virtual screening of large numbers of compounds against target protein binding sites has become an integral component of drug discovery workflows. This screening is often done by computationally docking ligands into a protein binding site of interest, but this has the drawback of a large number of poses that must be evaluated to obtain accurate estimates of protein-ligand binding affinity. We here introduce a fast pre-filtering method for ligand prioritization that is based on a set of machine learning models and uses simple pose-invariant physicochemical descriptors of the ligands and the protein binding pocket. Our method, Rapid Screening with Physicochemical Descriptors + machine learning (RASPD+), is trained on PDBbind data and achieves a regression performance that is better than that of the original RASPD method and traditional scoring functions on a range of different test sets without the need for generating ligand poses. Additionally, we use RASPD+ to identify molecular features important for binding affinity and assess the ability of RASPD+ to enrich active molecules from decoys.

Accurate predictions of aqueous solubility of drug molecules via the multilevel graph convolutional network (MGCN) and SchNet architectures

Deep learning based methods have been widely applied to predict various kinds of molecular properties in the pharmaceutical industry with increasingly more success. In this study, we propose two novel models for aqueous solubility predictions, based on the Multilevel Graph Convolutional Network (MGCN) and SchNet architectures, respectively. The advantage of the MGCN lies in the fact that it could extract the graph features of the target molecules directly from the (3D) structural information; therefore, it doesn't need to rely on a lot of intra-molecular descriptors to learn the features, which are of significance for accurate predictions of the molecular properties. The SchNet performs well in modelling the interatomic interactions inside a molecule, and such a deep learning architecture is also capable of extracting structural information and further predicting the related properties. The actual accuracy of these two novel approaches was systematically benchmarked with four different independent datasets. We found that both the MGCN and SchNet models performed well for aqueous solubility predictions. In the future, we believe such promising predictive models will be applicable to enhancing the efficiency of the screening, crystallization and delivery of drug molecules, essentially as a useful tool to promote the development of molecular pharmaceutics.