Protein language models are performant in structure-free virtual screening.

Hitherto virtual screening (VS) has been typically performed using a structure-based drug design paradigm. Such methods typically require the use of molecular docking on high-resolution three-dimensional structures of a target protein-a computationally-intensive and time-consuming exercise. This work demonstrates that by employing protein language models and molecular graphs as inputs to a novel graph-to-transformer cross-attention mechanism, a screening power comparable to state-of-the-art structure-based models can be achieved. The implications thereof include highly expedited VS due to the greatly reduced compute required to run this model, and the ability to perform early stages of computer-aided drug design in the complete absence of 3D protein structures.

Recognizing the power of machine learning and other computational methods to accelerate progress in small molecule targeting of RNA.

RNA structures regulate a wide range of processes in biology and disease, yet small molecule chemical probes or drugs that can modulate these functions are rare. Machine learning and other computational methods are well poised to fill gaps in knowledge and overcome the inherent challenges in RNA targeting, such as the dynamic nature of RNA and the difficulty of obtaining RNA high-resolution structures. Successful tools to date include principal component analysis, linear discriminate analysis, k-nearest neighbor, artificial neural networks, multiple linear regression, and many others. Employment of these tools has revealed critical factors for selective recognition in RNA:small molecule complexes, predictable differences in RNA- and protein-binding ligands, and quantitative structure activity relationships that allow the rational design of small molecules for a given RNA target. Herein we present our perspective on the value of using machine learning and other computation methods to advance RNA:small molecule targeting, including select examples and their validation as well as necessary and promising future directions that will be key to accelerate discoveries in this important field.

Using molecular embeddings in QSAR modeling: does it make a difference?

With the consolidation of deep learning in drug discovery, several novel algorithms for learning molecular representations have been proposed. Despite the interest of the community in developing new methods for learning molecular embeddings and their theoretical benefits, comparing molecular embeddings with each other and with traditional representations is not straightforward, which in turn hinders the process of choosing a suitable representation for Quantitative Structure-Activity Relationship (QSAR) modeling. A reason behind this issue is the difficulty of conducting a fair and thorough comparison of the different existing embedding approaches, which requires numerous experiments on various datasets and training scenarios. To close this gap, we reviewed the literature on methods for molecular embeddings and reproduced three unsupervised and two supervised molecular embedding techniques recently proposed in the literature. We compared these five methods concerning their performance in QSAR scenarios using different classification and regression datasets. We also compared these representations to traditional molecular representations, namely molecular descriptors and fingerprints. As opposed to the expected outcome, our experimental setup consisting of over $25 000$ trained models and statistical tests revealed that the predictive performance using molecular embeddings did not significantly surpass that of traditional representations. Although supervised embeddings yielded competitive results compared with those using traditional molecular representations, unsupervised embeddings tended to perform worse than traditional representations. Our results highlight the need for conducting a careful comparison and analysis of the different embedding techniques prior to using them in drug design tasks and motivate a discussion about the potential of molecular embeddings in computer-aided drug design.

Identification of two novel chemical classes of Autotaxin (ATX) inhibitors using Enalos Asclepios KNIME nodes.

Autotaxin is a secreted lysophospholipase D which is a member of the ectonucleotide pyrophosphatase/phosphodiesterase family converting extracellular lysophosphatidylcholine and other non-choline lysophospholipids, such as lysophosphatidylethanolamine and lysophosphatidylserine, to the lipid mediator lysophosphatidic acid. Autotaxin is implicated in various fibroproliferative diseases including interstitial lung diseases, such as idiopathic pulmonary fibrosis and hepatic fibrosis, as well as in cancer. In this study, we present an effort of identifying ATX inhibitors that bind to allosteric ATX binding sites using the Enalos Asclepios KNIME Node. All the available PDB crystal structures of ATX were collected, prepared, and aligned. Visual examination of these structures led to the identification of four crystal structures of human ATX co-crystallized with four known inhibitors. These inhibitors bind to five binding sites with five different binding modes. These five binding sites were thereafter used to virtually screen a compound library of 14,000 compounds to identify molecules that bind to allosteric sites. Based on the binding mode and interactions, the docking score, and the frequency that a compound comes up as a top-ranked among the five binding sites, 24 compounds were selected for in vitro testing. Finally, two compounds emerged with inhibitory activity against ATX in the low micromolar range, while their mode of inhibition and binding pattern were also studied. The two derivatives identified herein can serve as "hits" towards developing novel classes of ATX allosteric inhibitors.

Application of Transformers in Cheminformatics.

By accelerating time-consuming processes with high efficiency, computing has become an essential part of many modern chemical pipelines. Machine learning is a class of computing methods that can discover patterns within chemical data and utilize this knowledge for a wide variety of downstream tasks, such as property prediction or substance generation. The complex and diverse chemical space requires complex machine learning architectures with great learning power. Recently, learning models based on transformer architectures have revolutionized multiple domains of machine learning, including natural language processing and computer vision. Naturally, there have been ongoing endeavors in adopting these techniques to the chemical domain, resulting in a surge of publications within a short period. The diversity of chemical structures, use cases, and learning models necessitate a comprehensive summarization of existing works. In this paper, we review recent innovations in adapting transformers to solve learning problems in chemistry. Because chemical data is diverse and complex, we structure our discussion based on chemical representations. Specifically, we highlight the strengths and weaknesses of each representation, the current progress of adapting transformer architectures, and future directions.

Can Small Molecules Provide Clues on Disease Progression in Cerebrospinal Fluid from Mild Cognitive Impairment and Alzheimer's Disease Patients?

Alzheimer's disease (AD) is a complex and multifactorial neurodegenerative disease, which is currently diagnosed via clinical symptoms and nonspecific biomarkers (such as Aß1-42, t-Tau, and p-Tau) measured in cerebrospinal fluid (CSF), which alone do not provide sufficient insights into disease progression. In this pilot study, these biomarkers were complemented with small-molecule analysis using non-target high-resolution mass spectrometry coupled with liquid chromatography (LC) on the CSF of three groups: AD, mild cognitive impairment (MCI) due to AD, and a non-demented (ND) control group. An open-source cheminformatics pipeline based on MS-DIAL and patRoon was enhanced using CSF- and AD-specific suspect lists to assist in data interpretation. Chemical Similarity Enrichment Analysis revealed a significant increase of hydroxybutyrates in AD, including 3-hydroxybutanoic acid, which was found at higher levels in AD compared to MCI and ND. Furthermore, a highly sensitive target LC-MS method was used to quantify 35 bile acids (BAs) in the CSF, revealing several statistically significant differences including higher dehydrolithocholic acid levels and decreased conjugated BA levels in AD. This work provides several promising small-molecule hypotheses that could be used to help track the progression of AD in CSF samples.

Cross-column density functional theory-based quantitative structure-retention relationship model development powered by machine learning.

Quantitative structure-retention relationship (QSRR) modeling has emerged as an efficient alternative to predict analyte retention times using molecular descriptors. However, most reported QSRR models are column-specific, requiring separate models for each high-performance liquid chromatography (HPLC) system. This study evaluates the potential of machine learning (ML) algorithms and quantum mechanical (QM) descriptors to develop QSRR models that can predict retention times across three different reversed-phase HPLC columns under varying conditions. Four machine learning methods-partial least squares (PLS) regression, ridge regression (RR), random forest (RF), and gradient boosting (GB)-were compared on a dataset of 360 retention times for 15 aromatic analytes. Molecular descriptors were calculated using density functional theory (DFT). Column characteristics like particle size and pore size and experimental conditions like temperature and gradient time were additionally used as descriptors. Results showed that the GB-QSRR model demonstrated the best predictive performance, with Q2 of 0.989 and root mean square error of prediction (RMSEP) of 0.749 min on the test set. Feature analysis revealed that solvation energy (SE), HOMO-LUMO energy gap (∆E HOMO-LUMO), total dipole moment (Mtot), and global hardness (η) are among the most influential predictors for retention time prediction, indicating the significance of electrostatic interactions and hydrophobicity. Our findings underscore the efficiency of ensemble methods, GB and RF models employing non-linear learners, in capturing local variations in retention times across diverse experimental setups. This study emphasizes the potential of cross-column QSRR modeling and highlights the utility of ML models in optimizing chromatographic analysis.

Antimicrobial peptides designed by computational analysis of proteomes.

Antimicrobial peptides (AMPs) are promising cationic and amphipathic molecules to fight antibiotic resistance. To search for novel AMPs, we applied a computational strategy to identify peptide sequences within the organisms' proteome, including in-house developed software and artificial intelligence tools. After analyzing 150.450 proteins from eight proteomes of bacteria, plants, a protist, and a nematode, nine peptides were selected and modified to increase their antimicrobial potential. The 18 resulting peptides were validated by bioassays with four pathogenic bacterial species, one yeast species, and two cancer cell-lines. Fourteen of the 18 tested peptides were antimicrobial, with minimum inhibitory concentrations (MICs) values under 10 µM against at least three bacterial species; seven were active against Candida albicans with MICs values under 10 µM; six had a therapeutic index above 20; two peptides were active against A549 cells, and eight were active against MCF-7 cells under 30 µM. This study's most active antimicrobial peptides damage the bacterial cell membrane, including grooves, dents, membrane wrinkling, cell destruction, and leakage of cytoplasmic material. The results confirm that the proposed approach, which uses bioinformatic tools and rational modifications, is highly efficient and allows the discovery, with high accuracy, of potent AMPs encrypted in proteins.

Development of Drug Discovery Platforms Using Artificial Intelligence and Cheminformatics.

Recently, remarkable progress has been achieved in artificial intelligence (AI), including machine learning. Various AI models have been proposed for drug discovery, including the design of small molecules, activity prediction, and three-dimensional (3D) structure prediction of proteins. AI consists of diverse elements, including information retrieval and machine learning, and can be used in a wide range of drug discovery scenarios. In this review, we focused on AI for small-molecule drug discovery with respect to molecular design, activity prediction, and prediction of the binding poses of compounds to target molecules. We also discussed the applications of AI in academic drug discovery.

Machine Learning-Driven Classification of Urease Inhibitors Leveraging Physicochemical Properties as Effective Filter Criteria.

Urease, a pivotal enzyme in nitrogen metabolism, plays a crucial role in various microorganisms, including the pathogenic Helicobacter pylori. Inhibiting urease activity offers a promising approach to combating infections and associated ailments, such as chronic kidney diseases and gastric cancer. However, identifying potent urease inhibitors remains challenging due to resistance issues that hinder traditional approaches. Recently, machine learning (ML)-based models have demonstrated the ability to predict the bioactivity of molecules rapidly and effectively. In this study, we present ML models designed to predict urease inhibitors by leveraging essential physicochemical properties. The methodological approach involved constructing a dataset of urease inhibitors through an extensive literature search. Subsequently, these inhibitors were characterized based on physicochemical properties calculations. An exploratory data analysis was then conducted to identify and analyze critical features. Ultimately, 252 classification models were trained, utilizing a combination of seven ML algorithms, three attribute selection methods, and six different strategies for categorizing inhibitory activity. The investigation unveiled discernible trends distinguishing urease inhibitors from non-inhibitors. This differentiation enabled the identification of essential features that are crucial for precise classification. Through a comprehensive comparison of ML algorithms, tree-based methods like random forest, decision tree, and XGBoost exhibited superior performance. Additionally, incorporating the "chemical family type" attribute significantly enhanced model accuracy. Strategies involving a gray-zone categorization demonstrated marked improvements in predictive precision. This research underscores the transformative potential of ML in predicting urease inhibitors. The meticulous methodology outlined herein offers actionable insights for developing robust predictive models within biochemical systems.

Merging Counter-Propagation and Back-Propagation Algorithms: Overcoming the Limitations of Counter-Propagation Neural Network Models.

Artificial neural networks (ANNs) are nowadays applied as the most efficient methods in the majority of machine learning approaches, including data-driven modeling for assessment of the toxicity of chemicals. We developed a combined neural network methodology that can be used in the scope of new approach methodologies (NAMs) assessing chemical or drug toxicity. Here, we present QSAR models for predicting the physical and biochemical properties of molecules of three different datasets: aqueous solubility, acute fish toxicity toward fat head minnow, and bio-concentration factors. A novel neural network modeling method is developed by combining two neural network algorithms, namely, the counter-propagation modeling strategy (CP-ANN) with the back-propagation-of-errors algorithm (BPE-ANN). The advantage is a short training time, robustness, and good interpretability through the initial CP-ANN part, while the extension with BPE-ANN improves the precision of predictions in the range between minimal and maximal property values of the training data, regardless of the number of neurons in both neural networks, either CP-ANN or BPE-ANN.

MolOptimizer: A Molecular Optimization Toolkit for Fragment-Based Drug Design.

MolOptimizer is a user-friendly computational toolkit designed to streamline the hit-to-lead optimization process in drug discovery. MolOptimizer extracts features and trains machine learning models using a user-provided, labeled, and small-molecule dataset to accurately predict the binding values of new small molecules that share similar scaffolds with the target in focus. Hosted on the Azure web-based server, MolOptimizer emerges as a vital resource, accelerating the discovery and development of novel drug candidates with improved binding properties.

Expanding Predictive Capacities in Toxicology: Insights from Hackathon-Enhanced Data and Model Aggregation.

In the realm of predictive toxicology for small molecules, the applicability domain of QSAR models is often limited by the coverage of the chemical space in the training set. Consequently, classical models fail to provide reliable predictions for wide classes of molecules. However, the emergence of innovative data collection methods such as intensive hackathons have promise to quickly expand the available chemical space for model construction. Combined with algorithmic refinement methods, these tools can address the challenges of toxicity prediction, enhancing both the robustness and applicability of the corresponding models. This study aimed to investigate the roles of gradient boosting and strategic data aggregation in enhancing the predictivity ability of models for the toxicity of small organic molecules. We focused on evaluating the impact of incorporating fragment features and expanding the chemical space, facilitated by a comprehensive dataset procured in an open hackathon. We used gradient boosting techniques, accounting for critical features such as the structural fragments or functional groups often associated with manifestations of toxicity.

Synthesis and Cheminformatics-Directed Antibacterial Evaluation of Echinosulfonic Acid-Inspired Bis-Indole Alkaloids.

Synthetic efforts toward complex natural product (NP) scaffolds are useful ones, particularly those aimed at expanding their bioactive chemical space. Here, we utilised an orthogonal cheminformatics-based approach to predict the potential biological activities for a series of synthetic bis-indole alkaloids inspired by elusive sponge-derived NPs, echinosulfone A (1) and echinosulfonic acids A-D (2-5). Our work includes the first synthesis of desulfato-echinosulfonic acid C, an α-hydroxy bis(3'-indolyl) alkaloid (17), and its full NMR characterisation. This synthesis provides corroborating evidence for the structure revision of echinosulfonic acids A-C. Additionally, we demonstrate a robust synthetic strategy toward a diverse range of α-methine bis(3'-indolyl) acids and acetates (11-16) without the need for silica-based purification in either one or two steps. By integrating our synthetic library of bis-indoles with bioactivity data for 2048 marine indole alkaloids (reported up to the end of 2021), we analyzed their overlap with marine natural product chemical diversity. Notably, the C-6 dibrominated α-hydroxy bis(3'-indolyl) and α-methine bis(3'-indolyl) analogues (11, 14, and 17) were found to contain significant overlap with antibacterial C-6 dibrominated marine bis-indoles, guiding our biological evaluation. Validating the results of our cheminformatics analyses, the dibrominated α-methine bis(3'-indolyl) alkaloids (11, 12, 14, and 15) were found to exhibit antibacterial activities against methicillin-sensitive and -resistant Staphylococcus aureus. Further, while investigating other synthetic approaches toward bis-indole alkaloids, 16 incorrectly assigned synthetic α-hydroxy bis(3'-indolyl) alkaloids were identified. After careful analysis of their reported NMR data, and comparison with those obtained for the synthetic bis-indoles reported herein, all of the structures have been revised to α-methine bis(3'-indolyl) alkaloids.

Exploring Tau Fibril-Disaggregating and Antioxidating Molecules Binding to Membrane-Bound Amyloid Oligomers Using Machine Learning-Enhanced Docking and Molecular Dynamics.

Intracellular tau fibrils are sources of neurotoxicity and oxidative stress in Alzheimer's. Current drug discovery efforts have focused on molecules with tau fibril disaggregation and antioxidation functions. However, recent studies suggest that membrane-bound tau-containing oligomers (mTCOs), smaller and less ordered than tau fibrils, are neurotoxic in the early stage of Alzheimer's. Whether tau fibril-targeting molecules are effective against mTCOs is unknown. The binding of epigallocatechin-3-gallate (EGCG), CNS-11, and BHT-CNS-11 to in silico mTCOs and experimental tau fibrils was investigated using machine learning-enhanced docking and molecular dynamics simulations. EGCG and CNS-11 have tau fibril disaggregation functions, while the proposed BHT-CNS-11 has potential tau fibril disaggregation and antioxidation functions like EGCG. Our results suggest that the three molecules studied may also bind to mTCOs. The predicted binding probability of EGCG to mTCOs increases with the protein aggregate size. In contrast, the predicted probability of CNS-11 and BHT-CNS-11 binding to the dimeric mTCOs is higher than binding to the tetrameric mTCOs for the homo tau but not for the hetero tau-amylin oligomers. Our results also support the idea that anionic lipids may promote the binding of molecules to mTCOs. We conclude that tau fibril-disaggregating and antioxidating molecules may bind to mTCOs, and that mTCOs may also be useful targets for Alzheimer's drug design.

Cheminformatics-Guided Exploration of Synthetic Marine Natural Product-Inspired Brominated Indole-3-Glyoxylamides and Their Potentials for Drug Discovery.

Marine natural products (MNPs) continue to be tested primarily in cellular toxicity assays, both mammalian and microbial, despite most being inactive at concentrations relevant to drug discovery. These MNPs become missed opportunities and represent a wasteful use of precious bioresources. The use of cheminformatics aligned with published bioactivity data can provide insights to direct the choice of bioassays for the evaluation of new MNPs. Cheminformatics analysis of MNPs found in MarinLit (n = 39,730) up to the end of 2023 highlighted indol-3-yl-glyoxylamides (IGAs, n = 24) as a group of MNPs with no reported bioactivities. However, a recent review of synthetic IGAs highlighted these scaffolds as privileged structures with several compounds under clinical evaluation. Herein, we report the synthesis of a library of 32 MNP-inspired brominated IGAs (25-56) using a simple one-pot, multistep method affording access to these diverse chemical scaffolds. Directed by a meta-analysis of the biological activities reported for marine indole alkaloids (MIAs) and synthetic IGAs, the brominated IGAs 25-56 were examined for their potential bioactivities against the Parkinson's Disease amyloid protein alpha synuclein (α-syn), antiplasmodial activities against chloroquine-resistant (3D7) and sensitive (Dd2) parasite strains of Plasmodium falciparum, and inhibition of mammalian (chymotrypsin and elastase) and viral (SARS-CoV-2 3CLpro) proteases. All of the synthetic IGAs tested exhibited binding affinity to the amyloid protein α-syn, while some showed inhibitory activities against P. falciparum, and the proteases, SARS-CoV-2 3CLpro, and chymotrypsin. The cellular safety of the IGAs was examined against cancerous and non-cancerous human cell lines, with all of the compounds tested inactive, thereby validating cheminformatics and meta-analyses results. The findings presented herein expand our knowledge of marine IGA bioactive chemical space and advocate expanding the scope of biological assays routinely used to investigate NP bioactivities, specifically those more suitable for non-toxic compounds. By integrating cheminformatics tools and functional assays into NP biological testing workflows, we can aim to enhance the potential of NPs and their scaffolds for future drug discovery and development.

Phylogeny-Aware Chemoinformatic Analysis of Chemical Diversity in Lamiaceae Enables Iridoid Pathway Assembly and Discovery of Aucubin Synthase.

Countless reports describe the isolation and structural characterization of natural products, yet this information remains disconnected and underutilized. Using a cheminformatics approach, we leverage the reported observations of iridoid glucosides with the known phylogeny of a large iridoid producing plant family (Lamiaceae) to generate a set of biosynthetic pathways that best explain the extant iridoid chemical diversity. We developed a pathway reconstruction algorithm that connects iridoid reports via reactions and prunes this solution space by considering phylogenetic relationships between genera. We formulate a model that emulates the evolution of iridoid glucosides to create a synthetic data set, used to select the parameters that would best reconstruct the pathways, and apply them to the iridoid data set to generate pathway hypotheses. These computationally generated pathways were then used as the basis by which to select and screen biosynthetic enzyme candidates. Our model was successfully applied to discover a cytochrome P450 enzyme from Callicarpa americana that catalyzes the oxidation of bartsioside to aucubin, predicted by our model despite neither molecule having been observed in the genus. We also demonstrate aucubin synthase activity in orthologues of Vitex agnus-castus, and the outgroup Paulownia tomentosa, further strengthening the hypothesis, enabled by our model, that the reaction was present in the ancestral biosynthetic pathway. This is the first systematic hypothesis on the epi-iridoid glucosides biosynthesis in 25 years and sets the stage for streamlined work on the iridoid pathway. This work highlights how curation and computational analysis of widely available structural data can facilitate hypothesis-based gene discovery.

An automated method for graph-based chemical space exploration and transition state finding.

Algorithms that automatically explore the chemical space have been limited to chemical systems with a low number of atoms due to expensive involved quantum calculations and the large amount of possible reaction pathways. The method described here presents a novel solution to the problem of chemical exploration by generating reaction networks with heuristics based on chemical theory. First, a second version of the reaction network is determined through molecular graph transformations acting upon functional groups of the reacting. Only transformations that break two chemical bonds and form two new ones are considered, leading to a significant performance enhancement compared to previously presented algorithm. Second, energy barriers for this reaction network are estimated through quantum chemical calculations by a growing string method, which can also identify non-octet species missed during the previous step and further define the reaction network. The proposed algorithm has been successfully applied to five different chemical reactions, in all cases identifying the most important reaction pathways.

Cheminformatics and systems pharmacology approaches to unveil the potential plant bioactives to combat COVID-19.

COVID-19 was a global pandemic in the year 2020. Several treatment options failed to cure the disease. Thus, plant-based medicines are becoming a trend nowadays due to their less side effects. Bioactive chemicals from natural sources have been utilised for centuries as treatment options for a variety of ailments. To find out the potent bioactive compounds to counteract COVID-19, we use systems pharmacology and cheminformatics. They use the definitive data and predict the possible outcomes. In this study, we collected a total of 72 phytocompounds from the medicinally important plants such as Garcinia mangostana and Cinnamomum verum, of which 13 potential phytocompounds were identified to be active against the COVID-19 infection based on Swiss Target Prediction and compound target network analysis. These phytocompounds were annotated to identify the specific human receptor that targets COVID-19-specific genes such as MAPK8, MAPK14, ACE, CYP3A4, TLR4 and TYK2. Among these, compounds such as smeathxanthone A, demethylcalabaxanthone, mangostanol, trapezifolixanthone from Garcinia mangostana and camphene from C. verum were putatively target various COVID-19-related genes. Molecular docking results showed that smeathxanthone A and demethylcalabaxanthone exhibit increased binding efficiency towards the COVID-19-related receptor proteins. These compounds also showed efficient putative pharmacoactive properties than the commercial drugs ((R)-remdesivir, favipiravir and hydroxychloroquine) used to cure COVID-19. In conclusion, our study highlights the use of cheminformatics approach to unravel the potent and novel phytocompounds against COVID-19. These phytocompounds may be safer to use, more efficient and less harmful. This study highlights the value of natural products in the search for new drugs and identifies candidates with great promise.

Different molecular enumeration influences in deep learning: an example using aqueous solubility.

Aqueous solubility is the key property driving many chemical and biological phenomena and impacts experimental and computational attempts to assess those phenomena. Accurate prediction of solubility is essential and challenging, even with modern computational algorithms. Fingerprint-based, feature-based and molecular graph-based representations have all been used with different deep learning methods for aqueous solubility prediction. It has been clearly demonstrated that different molecular representations impact the model prediction and explainability. In this work, we reviewed different representations and also focused on using graph and line notations for modeling. In general, one canonical chemical structure is used to represent one molecule when computing its properties. We carefully examined the commonly used simplified molecular-input line-entry specification (SMILES) notation representing a single molecule and proposed to use the full enumerations in SMILES to achieve better accuracy. A convolutional neural network (CNN) was used. The full enumeration of SMILES can improve the presentation of a molecule and describe the molecule with all possible angles. This CNN model can be very robust when dealing with large datasets since no additional explicit chemistry knowledge is necessary to predict the solubility. Also, traditionally it is hard to use a neural network to explain the contribution of chemical substructures to a single property. We demonstrated the use of attention in the decoding network to detect the part of a molecule that is relevant to solubility, which can be used to explain the contribution from the CNN.