Discovery of Novel Inhibitors of a Critical Brain Enzyme Using a Homology Model and a Deep Convolutional Neural Network

Perspectives on current approaches to virtual screening in drug discovery

INTRODUCTION

For the past two decades, virtual screening (VS) has been an efficient hit finding approach for drug discovery. Today, billions of commercially accessible compounds are routinely screened, and many successful examples of VS have been reported. VS methods continue to evolve, including machine learning and physics-based methods.

AREAS COVERED

The authors examine recent examples of VS in drug discovery and discuss prospective hit finding results from the critical assessment of computational hit-finding experiments (CACHE) challenge. The authors also highlight the cost considerations and open-source options for conducting VS and examine chemical space coverage and library selections for VS.

EXPERT OPINION

The advancement of sophisticated VS approaches, including the use of machine learning techniques and increased computer resources as well as the ease of access to synthetically available chemical spaces, and commercial and open-source VS platforms allow for interrogating ultra-large libraries (ULL) of billions of molecules. An impressive number of prospective ULL VS campaigns have generated potent and structurally novel hits across many target classes. Nonetheless, many successful contemporary VS approaches still use considerably smaller focused libraries. This apparent dichotomy illustrates that VS is best conducted in a fit-for-purpose way choosing an appropriate chemical space. Better methods need to be developed to tackle more challenging targets.

Identification of potential pharmacological chaperones that selectively stabilize mutated Aspartoacylases in Canavan disease

Canavan disease is caused by mutations in the ASPA gene, leading to diminished catalytic activity of aspartoacylase in the brain. Clinical missense mutations are found throughout the enzyme structure, with many of these mutated enzymes having not only decreased activity but also compromised stability. High-throughput screening of a small molecule library has identified several compounds that significantly increase the thermal stability of the E285A mutant enzyme, the most predominant clinical mutation in Canavan disease, while having a negligible effect on the native enzyme. Based on the initial successes, some structural analogs of these initial hits were selected for further examination. Glutathione, NAAG and patulin were each confirmed to be competitive inhibitors, indicating the binding of these compounds at the dimer interface or near the active site of the E285A enzyme. The experimental results were theoretically examined with the help of the docking analysis method. The structure activity-guided optimization of these compounds can potentially lead to potential pharmacological chaperones that could alleviate the detrimental effect of ASPA mutations in Canavan patients.

Maurice M. Synthesis and Evaluation of 1,3-Disubstituted Imidazolidine-2,4,5-triones as Inhibitors of Pyruvate Carboxylase

Substituted imidazolidinetriones (IZTs) have been identified as potent inhibitors of pyruvate carboxylase (PC) through an in silico screening approach. Alkyl 2-(2,4,5-trioxo-3-substituted imidazolidin-1-yl)acetates (6i-6r) are the most potent of the series, with IC50 values between 3 and 12 μM, and several IZTs demonstrate high passive permeability across an artificial membrane. IZTs are mixed-type inhibitors with respect to pyruvate and noncompetitive with respect to ATP. This class of inhibitors appears to be selective for PC. Inhibitors in the IZT series do not inhibit the metalloenzymes human carbonic anhydrase II and matrix metalloprotease-12, and they do not inhibit the related biotin-dependent enzyme, guanidine carboxylase. Altogether, IZTs offer promise as PC inhibitors with potential downstream applications in cellular and in vivo systems.

Intracellular Allosteric Antagonist of the Olfactory Receptor OR51E2

Olfactory receptors are members of class A (rhodopsin-like) family of G protein-coupled receptors (GPCRs). Their expression and function have been increasingly studied in nonolfactory tissues, and many have been identified as potential therapeutic targets. In this manuscript, we focus on the discovery of novel ligands for the olfactory receptor family 51 subfamily E2 (OR51E2). We performed an artificial intelligence-based virtual drug screen of a ∼2.2 million small molecule library. Cell-based functional assay identified compound 80 (C80) as an antagonist and inverse agonist, and detailed pharmacological analysis revealed C80 acts as a negative allosteric modulator by significantly decreasing the agonist efficacy, while having a minimal effect on receptor affinity for agonist. C80 binds to an allosteric binding site formed by a network of nine residues localized in the intracellular parts of transmembrane domains 3, 5, 6, 7, and H8, which also partially overlaps with a G protein binding site. Mutational experiments of residues involved in C80 binding uncovered the significance of the C2406.37 position in blocking the activation-related conformational change and keeping the receptor in the inactive form. Our study provides a mechanistic understanding of the negative allosteric action of C80 on agonist-ctivated OR51E2. We believe the identification of the antagonist of OR51E2 will enable a multitude of studies aiming to determine the functional role of this receptor in specific biologic processes. SIGNIFICANCE STATEMENT: OR51E2 has been implicated in various biological processes, and its antagonists that can effectively modulate its activity have therapeutic potential. Here we report the discovery of a negative allosteric modulator of OR51E2 and provide a mechanistic understanding of its action. We demonstrate that this modulator has an inhibitory effect on the efficacy of the agonist for the receptor and reveal a network of nine residues that constitute its binding pocket, which also partially overlaps with the G protein binding site.

Comprehensive machine learning boosts structure-based virtual screening for PARP1 inhibitors

Poly ADP-ribose polymerase 1 (PARP1) is an attractive therapeutic target for cancer treatment. Machine-learning scoring functions constitute a promising approach to discovering novel PARP1 inhibitors. Cutting-edge PARP1-specific machine-learning scoring functions were investigated using semi-synthetic training data from docking activity-labelled molecules: known PARP1 inhibitors, hard-to-discriminate decoys property-matched to them with generative graph neural networks and confirmed inactives. We further made test sets harder by including only molecules dissimilar to those in the training set. Comprehensive analysis of these datasets using five supervised learning algorithms, and protein-ligand fingerprints extracted from docking poses and ligand only features revealed one highly predictive scoring function. This is the PARP1-specific support vector machine-based regressor, when employing PLEC fingerprints, which achieved a high Normalized Enrichment Factor at the top 1% on the hardest test set (NEF1% = 0.588, median of 10 repetitions), and was more predictive than any other investigated scoring function, especially the classical scoring function employed as baseline.

AI is a viable alternative to high throughput screening: a 318-target study

High throughput screening (HTS) is routinely used to identify bioactive small molecules. This requires physical compounds, which limits coverage of accessible chemical space. Computational approaches combined with vast on-demand chemical libraries can access far greater chemical space, provided that the predictive accuracy is sufficient to identify useful molecules. Through the largest and most diverse virtual HTS campaign reported to date, comprising 318 individual projects, we demonstrate that our AtomNet® convolutional neural network successfully finds novel hits across every major therapeutic area and protein class. We address historical limitations of computational screening by demonstrating success for target proteins without known binders, high-quality X-ray crystal structures, or manual cherry-picking of compounds. We show that the molecules selected by the AtomNet® model are novel drug-like scaffolds rather than minor modifications to known bioactive compounds. Our empirical results suggest that computational methods can substantially replace HTS as the first step of small-molecule drug discovery.

NAAG synthetase deficiency has only low influence on pathogenesis in a Canavan disease mouse model

Canavan disease (CD) is a leukodystrophy caused by mutations in the N-acetylaspartate (NAA) hydrolase aspartoacylase (ASPA). Inability to degrade NAA and its accumulation in the brain results in spongiform myelin degeneration. NAA is mainly synthesized by neurons, where it is also a precursor of the neuropeptide N-acetylaspartylglutamate (NAAG). Hydrolysis of this peptide by glutamate carboxypeptidases is an additional source of extracellular NAA besides the instant neuronal release of NAA. This study examines to what extent NAA released from NAAG contributes to NAA accumulation and pathogenesis in the brain of Aspanur7/nur7 mutant mice, an established model of CD. Towards this aim, Aspanur7/nur7 mice with additional deficiencies in NAAG synthetase genes Rimklb and/or Rimkla were generated. Loss of myelin in Aspanur7/nur7 mice was not significantly affected by Rimkla and Rimklb deficiency and there was also no obvious change in the extent of brain vacuolation. Astrogliosis was slightly reduced in the forebrain of Rimkla and Rimklb double deficient Aspanur7/nur7 mice. However, only minor improvements at the behavioral level were found. The brain NAA accumulation in CD mice was, however, not significantly reduced in the absence of NAAG synthesis. In summary, there was only a weak tendency towards reduced pathogenic symptoms in Aspanur7/nur7 mice deficient in NAAG synthesis. Therefore, we conclude that NAAG metabolism has little influence on NAA accumulation in Aspanur7/nur7 mice and development of pathological symptoms in CD.

The use of artificial intelligence in the treatment of rare diseases: A scoping review

With the increasing application of artificial intelligence (AI) in medicine and healthcare, AI technologies have the potential to improve the diagnosis, treatment, and prognosis of rare diseases. Presently, existing research predominantly focuses on the areas of diagnosis and prognosis, with relatively fewer studies dedicated to the domain of treatment. The purpose of this review is to systematically analyze the existing literature on the application of AI in the treatment of rare diseases. We searched three databases for related studies, and established criteria for the selection of retrieved articles. From the 407 unique articles identified across the three databases, 13 articles from 8 countries were selected, which investigated 10 different rare diseases. The most frequently studied rare disease group was rare neurologic diseases (n = 5/13, 38.46%). Among the four identified therapeutic domains, 7 articles (53.85%) focused on drug research, with 5 specifically focused on drug discovery (drug repurposing, the discovery of drug targets and small-molecule inhibitors), 1 on pre-clinical studies (drug interactions), and 1 on clinical studies (information strength assessment of clinical parameters). Across the selected 13 articles, we identified total 32 different algorithms, with random forest (RF) being the most commonly used (n = 4/32, 12.50%). The predominant purpose of AI in the treatment of rare diseases in these articles was to enhance the performance of analytical tasks (53.33%). The most common data source was database data (35.29%), with 5 of these studies being in the field of drug research, utilizing classic databases such as RCSB, PDB and NCBI. Additionally, 47.37% of the articles highlighted the existing challenge of data scarcity or small sample sizes.

Inactive-enriched machine-learning models exploiting patent data improve structure-based virtual screening for PDL1 dimerizers

INTRODUCTION

Small-molecule Programmable Cell Death Protein 1/Programmable Death-Ligand 1 (PD1/PDL1) inhibition via PDL1 dimerization has the potential to lead to inexpensive drugs with better cancer patient outcomes and milder side effects. However, this therapeutic approach has proven challenging, with only one PDL1 dimerizer reaching early clinical trials so far. There is hence a need for fast and accurate methods to develop alternative PDL1 dimerizers.

OBJECTIVES

We aim to show that structure-based virtual screening (SBVS) based on PDL1-specific machine-learning (ML) scoring functions (SFs) is a powerful drug design tool for detecting PD1/PDL1 inhibitors via PDL1 dimerization.

METHODS

By incorporating the latest MLSF advances, we generated and evaluated PDL1-specific MLSFs (classifiers and inactive-enriched regressors) on two demanding test sets.

RESULTS

60 PDL1-specific MLSFs (30 classifiers and 30 regressors) were generated. Our large-scale analysis provides highly predictive PDL1-specific MLSFs that benefitted from training with large volumes of docked inactives and enabling inactive-enriched regression.

CONCLUSION

PDL1-specific MLSFs strongly outperformed generic SFs of various types on this target and are released here without restrictions.

Artificial intelligence-powered discovery of small molecules inhibiting CTLA-4 in cancer

BACKGROUND/OBJECTIVES

Checkpoint inhibitors, which generate durable responses in many cancer patients, have revolutionized cancer immunotherapy. However, their therapeutic efficacy is limited, and immune-related adverse events are severe, especially for monoclonal antibody treatment directed against cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), which plays a pivotal role in preventing autoimmunity and fostering anticancer immunity by interacting with the B7 proteins CD80 and CD86. Small molecules impairing the CTLA-4/CD80 interaction have been developed; however, they directly target CD80, not CTLA-4.

SUBJECTS/METHODS

In this study, we performed artificial intelligence (AI)-powered virtual screening of approximately ten million compounds to identify those targeting CTLA-4. We validated the hits molecules with biochemical, biophysical, immunological, and experimental animal assays.

RESULTS

The primary hits obtained from the virtual screening were successfully validated in vitro and in vivo. We then optimized lead compounds and obtained inhibitors (inhibitory concentration, 1 micromole) that disrupted the CTLA-4/CD80 interaction without degrading CTLA-4.

CONCLUSIONS

Several compounds inhibited tumor development prophylactically and therapeutically in syngeneic and CTLA-4-humanized mice. Our findings support using AI-based frameworks to design small molecules targeting immune checkpoints for cancer therapy.

Computer-Selected Antiviral Compounds: Assessing In Vitro Efficacies against Rift Valley Fever Virus

Rift Valley fever is a zoonotic viral disease transmitted by mosquitoes, impacting both humans and livestock. Currently, there are no approved vaccines or antiviral treatments for humans. This study aimed to evaluate the in vitro efficacy of chemical compounds targeting the Gc fusion mechanism. These compounds were identified through virtual screening of millions of commercially available small molecules using a structure-based artificial intelligence bioactivity predictor. In our experiments, a pretreatment with small molecule compounds revealed that 3 out of 94 selected compounds effectively inhibited the replication of the Rift Valley fever virus MP-12 strain in Vero cells. As anticipated, these compounds did not impede viral RNA replication when administered three hours after infection. However, significant inhibition of viral RNA replication occurred upon viral entry when cells were pretreated with these small molecules. Furthermore, these compounds exhibited significant inhibition against Arumowot virus, another phlebovirus, while showing no antiviral effects on tick-borne bandaviruses. Our study validates AI-based virtual high throughput screening as a rational approach for identifying effective antiviral candidates for Rift Valley fever virus and other bunyaviruses.

Discovery of a novel cardiac-specific myosin modulator using artificial intelligence-based virtual screening

Direct modulation of cardiac myosin function has emerged as a therapeutic target for both heart disease and heart failure. However, the development of myosin-based therapeutics has been hampered by the lack of targeted in vitro screening assays. In this study we use Artificial Intelligence-based virtual high throughput screening (vHTS) to identify novel small molecule effectors of human β-cardiac myosin. We test the top scoring compounds from vHTS in biochemical counter-screens and identify a novel chemical scaffold called 'F10' as a cardiac-specific low-micromolar myosin inhibitor. Biochemical and biophysical characterization in both isolated proteins and muscle fibers show that F10 stabilizes both the biochemical (i.e. super-relaxed state) and structural (i.e. interacting heads motif) OFF state of cardiac myosin, and reduces force and left ventricular pressure development in isolated myofilaments and Langendorff-perfused hearts, respectively. F10 is a tunable scaffold for the further development of a novel class of myosin modulators.

A practical guide to machine-learning scoring for structure-based virtual screening

Structure-based virtual screening (SBVS) via docking has been used to discover active molecules for a range of therapeutic targets. Chemical and protein data sets that contain integrated bioactivity information have increased both in number and in size. Artificial intelligence and, more concretely, its machine-learning (ML) branch, including deep learning, have effectively exploited these data sets to build scoring functions (SFs) for SBVS against targets with an atomic-resolution 3D model (e.g., generated by X-ray crystallography or predicted by AlphaFold2). Often outperforming their generic and non-ML counterparts, target-specific ML-based SFs represent the state of the art for SBVS. Here, we present a comprehensive and user-friendly protocol to build and rigorously evaluate these new SFs for SBVS. This protocol is organized into four sections: (i) using a public benchmark of a given target to evaluate an existing generic SF; (ii) preparing experimental data for a target from public repositories; (iii) partitioning data into a training set and a test set for subsequent target-specific ML modeling; and (iv) generating and evaluating target-specific ML SFs by using the prepared training-test partitions. All necessary code and input/output data related to three example targets (acetylcholinesterase, HMG-CoA reductase, and peroxisome proliferator-activated receptor-α) are available at https://github.com/vktrannguyen/MLSF-protocol , can be run by using a single computer within 1 week and make use of easily accessible software/programs (e.g., Smina, CNN-Score, RF-Score-VS and DeepCoy) and web resources. Our aim is to provide practical guidance on how to augment training data to enhance SBVS performance, how to identify the most suitable supervised learning algorithm for a data set, and how to build an SF with the highest likelihood of discovering target-active molecules within a given compound library.

AI-powered discovery of a novel p53-Y220C reactivator

Introduction

The p53-Y220C mutation is one of the most common mutations that play a major role in cancer progression.

Methods

In this study, we applied artificial intelligence (AI)-powered virtual screening to identify small-molecule compounds that specifically restore the wild-type p53 conformation from p53-Y220C. From 10 million compounds, the AI algorithm selected a chemically diverse set of 83 high-scoring hits, which were subjected to several experimental assays using cell lines with different p53 mutations.

Results

We identified one compound, H3, that preferentially killed cells with the p53-Y220C mutation compared to cells with other p53 mutations. H3 increased the amount of folded mutant protein with wild-type p53 conformation, restored its transcriptional functions, and caused cell cycle arrest and apoptosis. Furthermore, H3 reduced tumorigenesis in a mouse xenograft model with p53-Y220C-positive cells.

Conclusion

AI enabled the discovery of the H3 compound that selectively reactivates the p53-Y220C mutant and inhibits tumor development in mice.

The Commoditization of AI for Molecule Design

Anyone involved in designing or finding molecules in the life sciences over the past few years has witnessed a dramatic change in how we now work due to the COVID-19 pandemic. Computational technologies like artificial intelligence (AI) seemed to become ubiquitous in 2020 and have been increasingly applied as scientists worked from home and were separated from the laboratory and their colleagues. This shift may be more permanent as the future of molecule design across different industries will increasingly require machine learning models for design and optimization of molecules as they become "designed by AI". AI and machine learning has essentially become a commodity within the pharmaceutical industry. This perspective will briefly describe our personal opinions of how machine learning has evolved and is being applied to model different molecule properties that crosses industries in their utility and ultimately suggests the potential for tight integration of AI into equipment and automated experimental pipelines. It will also describe how many groups have implemented generative models covering different architectures, for de novo design of molecules. We also highlight some of the companies at the forefront of using AI to demonstrate how machine learning has impacted and influenced our work. Finally, we will peer into the future and suggest some of the areas that represent the most interesting technologies that may shape the future of molecule design, highlighting how we can help increase the efficiency of the design-make-test cycle which is currently a major focus across industries.

Delta Machine Learning to Improve Scoring-Ranking-Screening Performances of Protein-Ligand Scoring Functions

Protein-ligand scoring functions are widely used in structure-based drug design for fast evaluation of protein-ligand interactions, and it is of strong interest to develop scoring functions with machine-learning approaches. In this work, by expanding the training set, developing physically meaningful features, employing our recently developed linear empirical scoring function Lin_F9 (Yang, C. J. Chem. Inf. Model. 2021, 61, 4630-4644) as the baseline, and applying extreme gradient boosting (XGBoost) with Δ-machine learning, we have further improved the robustness and applicability of machine-learning scoring functions. Besides the top performances for scoring-ranking-screening power tests of the CASF-2016 benchmark, the new scoring function Δ_{Lin_F9}XGB also achieves superior scoring and ranking performances in different structure types that mimic real docking applications. The scoring powers of Δ_{Lin_F9}XGB for locally optimized poses, flexible redocked poses, and ensemble docked poses of the CASF-2016 core set achieve Pearson's correlation coefficient (R) values of 0.853, 0.839, and 0.813, respectively. In addition, the large-scale docking-based virtual screening test on the LIT-PCBA data set demonstrates the reliability and robustness of Δ_{Lin_F9}XGB in virtual screening application. The Δ_{Lin_F9}XGB scoring function and its code are freely available on the web at (https://yzhang.hpc.nyu.edu/Delta_LinF9_XGB).

Discovery of Novel Trace Amine-Associated Receptor 5 (TAAR5) Antagonists Using a Deep Convolutional Neural Network

Trace amine-associated receptor 5 (TAAR5) is a G protein-coupled receptor that belongs to the TAARs family (TAAR1-TAAR9). TAAR5 is expressed in the olfactory epithelium and is responsible for sensing 3-methylamine (TMA). However, recent studies showed that TAAR5 is also expressed in the limbic brain regions and is involved in the regulation of emotional behaviour and adult neurogenesis, suggesting that TAAR5 antagonism may represent a novel therapeutic strategy for anxiety and depression. We used the AtomNet® model, the first deep learning neural network for structure-based drug discovery, to identify putative TAAR5 ligands and tested them in an in vitro BRET assay. We found two mTAAR5 antagonists with low to submicromolar activity that are able to inhibit the cAMP production induced by TMA. Moreover, these two compounds also inhibited the mTAAR5 downstream signalling, such as the phosphorylation of CREB and ERK. These two hits exhibit drug-like properties and could be used to further develop more potent TAAR5 ligands with putative anxiolytic and antidepressant activity.

AtomNet PoseRanker: Enriching Ligand Pose Quality for Dynamic Proteins in Virtual High-Throughput Screens

Structure-based, virtual High-Throughput Screening (vHTS) methods for predicting ligand activity in drug discovery are important when there are no or relatively few known compounds that interact with a therapeutic target of interest. State-of-the-art computational vHTS necessarily relies on effective methods for pose sampling and docking and generating an accurate affinity score from the docked poses. However, proteins are dynamic; in vivo ligands bind to a conformational ensemble. In silico docking to the single conformation represented by a crystal structure can adversely affect the pose quality. Here, we introduce AtomNet PoseRanker (ANPR), a graph convolutional network trained to identify and rerank crystal-like ligand poses from a sampled ensemble of protein conformations and ligand poses. In contrast to conventional vHTS methods that incorporate receptor flexibility, a deep learning approach can internalize valid cognate and noncognate binding modes corresponding to distinct receptor conformations, thereby learning to infer and account for receptor flexibility even on single conformations. ANPR significantly enriched pose quality in docking to cognate and noncognate receptors of the PDBbind v2019 data set. Improved pose rankings that better represent experimentally observed ligand binding modes improve hit rates in vHTS campaigns and thereby advance computational drug discovery, especially for novel therapeutic targets or novel binding sites.

High Throughput Screening Cascade To Identify Human Aspartate N-Acetyltransferase (ANAT) Inhibitors for Canavan Disease

Canavan disease (CD) is a progressive, fatal neurological disorder that begins in infancy resulting from a mutation in aspartoacyclase (ASPA), an enzyme that catalyzes the deacetylation of N-acetyl aspartate (NAA) into acetate and aspartate. Increased NAA levels in the brains of affected children are one of the hallmarks of CD. Interestingly, genetic deletion of N-acetyltransferase-8-like (NAT8L), which encodes aspartate N-aceyltransferase (ANAT), an enzyme responsible for the synthesis of NAA from l-aspartate and acetyl-CoA, leads to normalization of NAA levels and improvement of symptoms in several genetically engineered mouse models of CD. Therefore, pharmacological inhibition of ANAT presents a promising therapeutic strategy for treating CD. Currently, however, there are no clinically viable ANAT inhibitors. Herein we describe the development of fluorescence-based high throughput screening (HTS) and radioactive-based orthogonal assays using recombinant human ANAT expressed in E. coli. In the fluorescence-based assay, ANAT activity was linear with respect to time of incubation up to 30 min and protein concentration up to 97.5 ng/μL with K_m values for l-aspartate and acetyl-CoA of 237 μM and 11 μM, respectively. Using this optimized assay, we conducted a pilot screening of a 10 000-compound library. Hits from the fluorescence-based assay were subjected to an orthogonal radioactive-based assay using L-[U-¹⁴C] aspartate as a substrate. Two compounds were confirmed to have dose-dependent inhibition in both assays. Inhibitory kinetics studies of the most potent compound revealed an uncompetitive inhibitory mechanism with respect to l-aspartate and a noncompetitive inhibitory mechanism against acetyl-CoA. The screening cascade developed herein will enable large-scale compound library screening to identify novel ANAT inhibitors as leads for further medicinal chemistry optimization.

The Communication of Hit Quality Using Natural History Visualizations (NHVs)

High-throughput screening (HTS) often yields a list of compounds that requires prioritization before further work is performed. Prioritization criteria typically include activity, selectivity, physicochemical properties, and other absolute or calculated measurements of compound "value." One critical method of compound prioritization is often not discussed in published accounts of HTS. We have referred to this oft-overlooked metric as "compound natural history." These natural histories are observational evaluations of how a compound has been reported in the historical literature or compound databases. The purpose of this work was to develop a useful natural history visualization (NHV) that could form a standard, important part of hit reporting and evaluation. In this case report, we propose an efficient and effective NHV that will assist in the prioritization of active compounds and demonstrate its utility using a retrospective analysis of reported hits. We propose that this method of compound natural history evaluation be adopted in HTS triage and become an integral component of published reports of HTS outcomes.

Computational representations of protein-ligand interfaces for structure-based virtual screening

Introduction: Structure-based virtual screening (SBVS) is an essential strategy for hit identification. SBVS primarily uses molecular docking, which exploits the protein-ligand binding mode and associated affinity score for compound ranking. Previous studies have shown that computational representation of protein-ligand interfaces and the later establishment of machine learning models are efficacious in improving the accuracy of SBVS.Areas covered: The authors review the computational methods for representing protein-ligand interfaces, which include the traditional ones that use deliberately designed fingerprints and descriptors and the more recent methods that automatically extract features with deep learning. The effects of these methods on the performance of machine learning models are briefly discussed. Additionally, case studies that applied various computational representations to machine learning are cited with remarks.Expert opinion: It has become a trend to extract binding features automatically by deep learning, which uses a completely end-to-end representation. However, there is still plenty of scope for improvement . The interpretability of deep-learning models, the organization of data management, the quantity and quality of available data, and the optimization of hyperparameters could impact the accuracy of feature extraction. In addition, other important structural factors such as water molecules and protein flexibility should be considered.

Recent progress on the prospective application of machine learning to structure-based virtual screening

As more bioactivity and protein structure data become available, scoring functions (SFs) using machine learning (ML) to leverage these data sets continue to gain further accuracy and broader applicability. Advances in our understanding of the optimal ways to train and evaluate these ML-based SFs have introduced further improvements. One of these advances is how to select the most suitable decoys (molecules assumed inactive) to train or test an ML-based SF on a given target. We also review the latest applications of ML-based SFs for prospective structure-based virtual screening (SBVS), with a focus on the observed improvement over those using classical SFs. Finally, we provide recommendations for future prospective SBVS studies based on the findings of recent methodological studies.

Big Techs and startups in pharmaceutical R&D - A 2020 perspective on artificial intelligence

We investigated what kind of artificial intelligence (AI) technologies are utilized in pharmaceutical research and development (R&D) and which sources of AI-related competencies can be leveraged by pharmaceutical companies. First, we found that machine learning (ML) is the dominating AI technology currently used in pharmaceutical R&D. Second, both Big Techs and AI startups are competent knowledge bases for AI applications. Big Techs have long-lasting experience in the digital field and offer more general IT solutions to support pharmaceutical companies in cloud computing, health monitoring, diagnostics or clinical trial management, whereas startups can provide more specific AI services to address special issues in the drug-discovery space.

Machine learning techniques applied to the drug design and discovery of new antivirals: a brief look over the past decade

Introduction: Drug design and discovery of new antivirals will always be extremely important in medicinal chemistry, taking into account known and new viral diseases that are yet to come. Although machine learning (ML) have shown to improve predictions on the biological potential of chemicals and accelerate the discovery of drugs over the past decade, new methods and their combinations have improved their performance and established promising perspectives regarding ML in the search for new antivirals.Areas covered: The authors consider some interesting areas that deal with different ML techniques applied to antivirals. Recent innovative studies on ML and antivirals were selected and analyzed in detail. Also, the authors provide a brief look at the past to the present to detect advances and bottlenecks in the area.Expert opinion: From classical ML techniques, it was possible to boost the searches for antivirals. However, from the emergence of new algorithms and the improvement in old approaches, promising results will be achieved every day, as we have observed in the case of SARS-CoV-2. Recent experience has shown that it is possible to use ML to discover new antiviral candidates from virtual screening and drug repurposing.

Engineering of a critical membrane-anchored enzyme for high solubility and catalytic activity

Membrane-associated proteins carry out a wide range of essential cellular functions but the structural characterization needed to understand these functions is dramatically underrepresented in the Protein Data Bank. Producing a soluble, stable and active form of a membrane-associated protein presents formidable challenges, as evidenced by the variety of approaches that have been attempted with a multitude of different membrane proteins to achieve this goal. Aspartate N-acetyltransferase (ANAT) is a membrane-anchored enzyme that performs a critical function, the synthesis of N-acetyl-l-aspartate (NAA), the second most abundant amino acid in the brain. This amino acid is a precursor for a neurotransmitter, and alterations in brain NAA levels have been implicated as a causative effect in Canavan disease and has been suggested to be involved in other neurological disorders. Numerous prior attempts have failed to produce a soluble form of ANAT that is amenable for functional and structural investigations. Through the application of a range of different approaches, including fusion partner constructs, linker modifications, membrane-anchor modifications, and domain truncations, a highly soluble, stable and fully active form of ANAT has now been obtained. Producing this modified enzyme form will accelerate studies aimed at structural characterization and structure-guided inhibitor development.

Generating Property-Matched Decoy Molecules Using Deep Learning

MOTIVATION

An essential step in the development of virtual screening methods is the use of established sets of actives and decoys for benchmarking and training. However, the decoy molecules in commonly used sets are biased meaning that methods often exploit these biases to separate actives and decoys, and do not necessarily learn to perform molecular recognition. This fundamental issue prevents generalisation and hinders virtual screening method development.

RESULTS

We have developed a deep learning method (DeepCoy) that generates decoys to a user's preferred specification in order to remove such biases or construct sets with a defined bias.We validated DeepCoy using two established benchmarks, DUD-E and DEKOIS 2.0. For all 102 DUD-E targets and 80 of the 81 DEKOIS 2.0 targets, our generated decoy molecules more closely matched the active molecules' physicochemical properties while introducing no discernible additional risk of false negatives. The DeepCoy decoys improved the Deviation from Optimal Embedding (DOE) score by an average of 81% and 66%, respectively, decreasing from 0.166 to 0.032 for DUD-E and from 0.109 to 0.038 for DEKOIS 2.0. Further, the generated decoys are harder to distinguish than the original decoy molecules via docking with Autodock Vina, with virtual screening performance falling from an AUC ROC of 0.70 to 0.63.

AVAILABILITY AND IMPLEMENTATION

The code is available at https://github.com/oxpig/DeepCoy. Generated molecules can be downloaded from http://opig.stats.ox.ac.uk/resources.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.