Chemoinformatic Classification Methods and their Applicability Domain

APBIO: bioactive profiling of air pollutants through inferred bioactivity signatures and prediction of novel target interactions

More sophisticated representations of compounds attempt to incorporate not only information on the structure and physicochemical properties of molecules, but also knowledge about their biological traits, leading to the so-called bioactivity profile. The bioactive profiling of air pollutants is challenging and crucial, as their biological activity and toxicological effects have not been deeply investigated yet, and further exploration could shed light on the impact of air pollution on complex disorders. Therefore, a biological signature that simultaneously captures the chemistry and the biology of small molecules may be beneficial in predicting the behaviour of such ligands towards a protein target. Moreover, the interactivity between biological entities can be represented through combined feature vectors that can be given as input to a machine learning (ML) model to capture the underlying interaction. To this end, we propose a chemogenomic approach, called Air Pollutant Bioactivity (APBIO), which integrates compound bioactivity signatures and target sequence descriptors to train ML classifiers subsequently used to predict potential compound-target interactions (CTIs). We report the performances of the proposed methodology and, via external validation sets, demonstrate its outperformance compared to existing molecular representations in terms of model generalizability. We have also developed a publicly available Streamlit application for APBIO at ap-bio.streamlit.app, allowing users to predict associations between investigated compounds and protein targets.Scientific contributionWe derived ex novo bioactivity signatures for air pollutant molecules to capture their biological behaviour and associations with protein targets. The proposed chemogenomic methodology enables the prediction of novel CTIs for known or similar compounds and targets through well-established and efficient ML models, deepening our insight into the molecular interactions and mechanisms that may have a deleterious impact on human biological systems.

Inverse Design of Singlet-Fission Materials with Uncertainty-Controlled Genetic Optimization

Singlet fission has shown potential for boosting the efficiency of solar cells, but the scarcity of suitable molecular materials hinders its implementation. We introduce an uncertainty-controlled genetic algorithm (ucGA) based on ensemble machine learning predictions from different molecular representations that concurrently optimizes excited state energies, synthesizability, and exciton size for the discovery of singlet fission materials. The ucGA allows us to efficiently explore the chemical space spanned by the reFORMED fragment database, which consists of 45,000 cores and 5,000 substituents derived from crystallographic structures assembled in the FORMED repository. Running the ucGA in an exploitative setup performs local optimization on variations of known singlet fission scaffolds, such as acenes. In an explorative mode, hitherto unknown candidates displaying excellent excited state properties for singlet fission are generated. We suggest a class of heteroatom-rich mesoionic compounds as acceptors for charge-transfer mediated singlet fission. When included in larger donor-acceptor systems, these units exhibit localization of the triplet state, favorable diradicaloid character and suitable triplet energies for exciton injection into semiconductor solar cells.

UNIQUE: A Framework for Uncertainty Quantification Benchmarking

Machine learning (ML) models have become key in decision-making for many disciplines, including drug discovery and medicinal chemistry. ML models are generally evaluated prior to their usage in high-stakes decisions, such as compound synthesis or experimental testing. However, no ML model is robust or predictive in all real-world scenarios. Therefore, uncertainty quantification (UQ) in ML predictions has gained importance in recent years. Many investigations have focused on developing methodologies that provide accurate uncertainty estimates for ML-based predictions. Unfortunately, there is no UQ strategy that consistently provides robust estimates about model's applicability on new samples. Depending on the dataset, prediction task, and algorithm, accurate uncertainty estimations might be unfeasible to obtain. Moreover, the optimum UQ metric also varies across applications, and previous investigations have shown a lack of consistency across benchmarks. Herein, the UNIQUE (UNcertaInty QUantification bEnchmarking) framework is introduced to facilitate a comparison of UQ strategies in ML-based predictions. This Python library unifies the benchmarking of multiple UQ metrics, including the calculation of nonstandard UQ metrics (combining information from the dataset and model), and provides a comprehensive evaluation. In this framework, UQ metrics are evaluated for different application scenarios, e.g., eliminating the predictions with the lowest confidence or obtaining a reliable uncertainty estimate for an acquisition function. Taken together, this library will help to standardize UQ investigations and evaluate new methodologies.

Machine Learning-Based In Silico Prediction of the Inhibitory Activity of Chemical Substances Against Rat and Human Cytochrome P450s

The prediction of cytochrome P450 inhibition by a computational (quantitative) structure-activity relationship approach using chemical structure information and machine learning would be useful for toxicity research as a simple and rapid in silico tool. However, there are few in silico models focusing on the species differences between rat and human in the P450s inhibition. This study aimed to establish in silico models to classify chemical substances as inhibitors or non-inhibitors of various rat and human P450s, using only molecular descriptors. Using the in-house test results from our in vitro experiments, we used 326 substances for model construction and internal validation data. Apart from the 326 substances, 60 substances were used as external validation data set. We focused on seven rat P450s (CYP1A1, CYP1A2, CYP2B1, CYP2C6, CYP2D1, CYP2E1, and CYP3A2) and 11 human P450s (CYP1A1, CYP1A2, CYP1B1, CYP2A6, CYP2B6, CYP2C8, CYP2C9, CYP2C19, CYP2D6, CYP2E1, and CYP3A4). Most of the models established using XGBoost showed an area under the receiver operating characteristic curve (ROC-AUC) of 0.8 or more in the internal validation. When we set an applicability domain for the models and confirmed their generalization performance through external validation, most of the models showed an ROC-AUC of 0.7 or more. Interestingly, for CYP1A1 and CYP1A2, we discovered that a human P450 inhibitory activity model can predict rat P450 inhibitory activity and vice versa. These models are the first attempts to predict inhibitory activity against a wide variety of P450s in both rats and humans using chemical structure information. Our experimental results and in silico models would be helpful to support information for species similarities and differences in chemical-induced toxicity.

Quantitative structure-activity relationships of chemical bioactivity toward proteins associated with molecular initiating events of organ-specific toxicity

The adverse outcome pathway (AOP) concept has gained attention as a way to explore the mechanism of chemical toxicity. In this study, quantitative structure-activity relationship (QSAR) models were developed to predict compound activity toward protein targets relevant to molecular initiating events (MIE) upstream of organ-specific toxicities, namely liver steatosis, cholestasis, nephrotoxicity, neural tube closure defects, and cognitive functional defects. Utilizing bioactivity data from the ChEMBL 33 database, various machine learning algorithms, chemical features and methods to assess prediction reliability were compared and applied to develop robust models to predict compound activity. The results demonstrate high predictive performance across multiple targets, with balanced accuracy exceeding 0.80 for the majority of models. Furthermore, stability checks confirmed the consistency of predictive performance across multiple training-test splits. The results obtained by using QSAR predictions to identify known markers of adversities highlighted the utility of the models for risk assessment and for prioritizing compounds for further experimental evaluation.Scientific contributionThe work describes the development of QSAR models as tools for screening chemicals with potential systemic toxicity, thus contributing to resource savings and providing indications for further better-targeted testing. This study provides advances in the field of computational modeling of MIEs and information from AOP which is still relatively young and unexplored. The comprehensive modeling procedure is highly generalizable, and offers a robust framework for predicting a wide range of toxicological endpoints.

Data-centric challenges with the application and adoption of artificial intelligence for drug discovery

INTRODUCTION

Artificial intelligence (AI) is exhibiting tremendous potential to reduce the massive costs and long timescales of drug discovery. There are however important challenges currently limiting the impact and scope of AI models.

AREAS COVERED

In this perspective, the authors discuss a range of data issues (bias, inconsistency, skewness, irrelevance, small size, high dimensionality), how they challenge AI models, and which issue-specific mitigations have been effective. Next, they point out the challenges faced by uncertainty quantification techniques aimed at enhancing and trusting the predictions from these AI models. They also discuss how conceptual errors, unrealistic benchmarks and performance misestimation can confound the evaluation of models and thus their development. Lastly, the authors explain how human bias, whether from AI experts or drug discovery experts, constitutes another challenge that can be alleviated by gaining more prospective experience.

EXPERT OPINION

AI models are often developed to excel on retrospective benchmarks unlikely to anticipate their prospective performance. As a result, only a few of these models are ever reported to have prospective value (e.g. by discovering potent and innovative drug leads for a therapeutic target). The authors have discussed what can go wrong in practice with AI for drug discovery. The authors hope that this will help inform the decisions of editors, funders investors, and researchers working in this area.

Insights into nanoparticle toxicity against aquatic organisms using multivariate regression, read-across, and ML algorithms: Predictive models for Daphnia magna and Danio rerio

The production of nanoparticles (NPs) has recently become more prevalent owing to their numerous applications in the fast-growing nanotechnology industry. Although nanoparticles have growing applications, there is a significant concern over their environmental impact due to their inevitable release into the environment. With the increasing risk to aquatic organisms, D. magna and zebrafish (Danio rerio) have been preferred as important freshwater model organisms for risk assessment and ecotoxicological studies on metal oxide-based nanoparticles (MeOxNPs) in aquatic environments. It is unfeasible to assess the risks associated with every single NP through in vivo or in vitro experiments. As an alternative, in silico approaches are employed to evaluate the NP toxicity. To evaluate such performance, we have collected data from databases and literature reviews to develop models based on multivariate regression, read-across approach (RA), and machine learning (ML) algorithms following the principles of OECD (Organization for Economic Cooperation and Development) for QSAR modeling. This work has aimed to investigate which features are important drivers of nanotoxicity in D. magna and Danio rerio using simple periodic table-derived descriptors. Further, we have examined the effectiveness of read-across-derived similarity measures compared to traditional QSAR models. The results obtained from model 1 infers that nanoparticles' size, the number of metals, the core environment of the metal present in the metal oxide, and the oxidation number of the metal play a key role in the final expression of toxicity of nanoparticles to D. magna. On the other hand, the presence of higher molecular weight, the core of the metal, and the presence of oxygen influence the enzyme inhibition activity. The enzyme inhibition is correlated with the ability of zebrafish embryos to hatch, and therefore, the inhibition of ZHE1 seems to be the factor driving hatch delay. The study emphasized the importance of developing transferable, reproducible, and easily interpretable models for the early identification of nanoparticle features contributing to aquatic toxicity.

Do Molecular Fingerprints Identify Diverse Active Drugs in Large-Scale Virtual Screening? (No)

Computational approaches for small-molecule drug discovery now regularly scale to the consideration of libraries containing billions of candidate small molecules. One promising approach to increased the speed of evaluating billion-molecule libraries is to develop succinct representations of each molecule that enable the rapid identification of molecules with similar properties. Molecular fingerprints are thought to provide a mechanism for producing such representations. Here, we explore the utility of commonly used fingerprints in the context of predicting similar molecular activity. We show that fingerprint similarity provides little discriminative power between active and inactive molecules for a target protein based on a known active-while they may sometimes provide some enrichment for active molecules in a drug screen, a screened data set will still be dominated by inactive molecules. We also demonstrate that high-similarity actives appear to share a scaffold with the query active, meaning that they could more easily be identified by structural enumeration. Furthermore, even when limited to only active molecules, fingerprint similarity values do not correlate with compound potency. In sum, these results highlight the need for a new wave of molecular representations that will improve the capacity to detect biologically active molecules based on their similarity to other such molecules.

Trust me if you can: a survey on reliability and interpretability of machine learning approaches for drug sensitivity prediction in cancer

With the ever-increasing number of artificial intelligence (AI) systems, mitigating risks associated with their use has become one of the most urgent scientific and societal issues. To this end, the European Union passed the EU AI Act, proposing solution strategies that can be summarized under the umbrella term trustworthiness. In anti-cancer drug sensitivity prediction, machine learning (ML) methods are developed for application in medical decision support systems, which require an extraordinary level of trustworthiness. This review offers an overview of the ML landscape of methods for anti-cancer drug sensitivity prediction, including a brief introduction to the four major ML realms (supervised, unsupervised, semi-supervised, and reinforcement learning). In particular, we address the question to what extent trustworthiness-related properties, more specifically, interpretability and reliability, have been incorporated into anti-cancer drug sensitivity prediction methods over the previous decade. In total, we analyzed 36 papers with approaches for anti-cancer drug sensitivity prediction. Our results indicate that the need for reliability has hardly been addressed so far. Interpretability, on the other hand, has often been considered for model development. However, the concept is rather used intuitively, lacking clear definitions. Thus, we propose an easily extensible taxonomy for interpretability, unifying all prevalent connotations explicitly or implicitly used within the field.

Chemoinformatic regression methods and their applicability domain

The growing interest in chemoinformatic model uncertainty calls for a summary of the most widely used regression techniques and how to estimate their reliability. Regression models learn a mapping from the space of explanatory variables to the space of continuous output values. Among other limitations, the predictive performance of the model is restricted by the training data used for model fitting. Identification of unusual objects by outlier detection methods can improve model performance. Additionally, proper model evaluation necessitates defining the limitations of the model, often called the applicability domain. Comparable to certain classifiers, some regression techniques come with built-in methods or augmentations to quantify their (un)certainty, while others rely on generic procedures. The theoretical background of their working principles and how to deduce specific and general definitions for their domain of applicability shall be explained.

Building species trait-specific nano-QSARs: Model stacking, navigating model uncertainties and limitations, and the effect of dataset size

A strong need exists for broadly applicable nano-QSARs, capable of predicting toxicological outcomes towards untested species and nanomaterials, under different environmental conditions. Existing nano-QSARs are generally limited to only a few species but the inclusion of species characteristics into models can aid in making them applicable to multiple species, even when toxicity data is not available for biological species. Species traits were used to create classification- and regression machine learning models to predict acute toxicity towards aquatic species for metallic nanomaterials. Afterwards, the individual classification- and regression models were stacked into a meta-model to improve performance. Additionally, the uncertainty and limitations of the models were assessed in detail (beyond the OECD principles) and it was investigated whether models would benefit from the addition of more data. Results showed a significant improvement in model performance following model stacking. Investigation of model uncertainties and limitations highlighted the discrepancy between the applicability domain and accuracy of predictions. Data points outside of the assessed chemical space did not have higher likelihoods of generating inadequate predictions or vice versa. It is therefore concluded that the applicability domain does not give complete insight into the uncertainty of predictions and instead the generation of prediction intervals can help in this regard. Furthermore, results indicated that an increase of the dataset size did not improve model performance. This implies that larger dataset sizes may not necessarily improve model performance while in turn also meaning that large datasets are not necessarily required for prediction of acute toxicity with nano-QSARs.

Andrographolide: A Diterpenoid from Cymbopogon schoenanthus Identified as a New Hit Compound against Trypanosoma cruzi Using Machine Learning and Experimental Approaches

American Trypanosomiasis, also known as Chagas disease, is caused by the protozoan Trypanosoma cruzi and exhibits limited options for treatment. Natural products offer various structurally complex metabolites with biological activities, including those with anti-T. cruzi potential. The discovery and development of prototypes based on natural products frequently display multiple phases that could be facilitated by machine learning techniques to provide a fast and efficient method for selecting new hit candidates. Using Random Forest and k-Nearest Neighbors, two models were constructed to predict the biological activity of natural products from plants against intracellular amastigotes of T. cruzi. The diterpenoid andrographolide was identified from a virtual screening as a promising hit compound. Hereafter, it was isolated from Cymbopogon schoenanthus and chemically characterized by spectral data analysis. Andrographolide was evaluated against trypomastigote and amastigote forms of T. cruzi, showing IC50 values of 29.4 and 2.9 μM, respectively, while the standard drug benznidazole displayed IC50 values of 17.7 and 5.0 μM, respectively. Additionally, the isolated compound exhibited a reduced cytotoxicity (CC50 = 92.8 μM) against mammalian cells and afforded a selectivity index (SI) of 32, similar to that of benznidazole (SI = 39). From the in silico analyses, we can conclude that andrographolide fulfills many requirements implemented by DNDi to be a hit compound. Therefore, this work successfully obtained machine learning models capable of predicting the activity of compounds against intracellular forms of T. cruzi.

Prediction of compound-target interaction using several artificial intelligence algorithms and comparison with a consensus-based strategy

For understanding a chemical compound's mechanism of action and its side effects, as well as for drug discovery, it is crucial to predict its possible protein targets. This study examines 15 developed target-centric models (TCM) employing different molecular descriptions and machine learning algorithms. They were contrasted with 17 third-party models implemented as web tools (WTCM). In both sets of models, consensus strategies were implemented as potential improvement over individual predictions. The findings indicate that TCM reach f1-score values greater than 0.8. Comparing both approaches, the best TCM achieves values of 0.75, 0.61, 0.25 and 0.38 for true positive/negative rates (TPR, TNR) and false negative/positive rates (FNR, FPR); outperforming the best WTCM. Moreover, the consensus strategy proves to have the most relevant results in the top 20 % of target profiles. TCM consensus reach TPR and FNR values of 0.98 and 0; while on WTCM reach values of 0.75 and 0.24. The implemented computational tool with the TCM and their consensus strategy at: https://bioquimio.udla.edu.ec/tidentification01/ . Scientific Contribution: We compare and discuss the performances of 17 public compound-target interaction prediction models and 15 new constructions. We also explore a compound-target interaction prioritization strategy using a consensus approach, and we analyzed the challenging involved in interactions modeling.

Rethinking the applicability domain analysis in QSAR models

Notwithstanding the wide adoption of the OECD principles (or best practices) for QSAR modeling, disparities between in silico predictions and experimental results are frequent, suggesting that model predictions are often too optimistic. Of these OECD principles, the applicability domain (AD) estimation has been recognized in several reports in the literature to be one of the most challenging, implying that the actual reliability measures of model predictions are often unreliable. Applying tree-based error analysis workflows on 5 QSAR models reported in the literature and available in the QsarDB repository, i.e., androgen receptor bioactivity (agonists, antagonists, and binders, respectively) and membrane permeability (highest membrane permeability and the intrinsic permeability), we demonstrate that predictions erroneously tagged as reliable (AD prediction errors) overwhelmingly correspond to instances in subspaces (cohorts) with the highest prediction error rates, highlighting the inhomogeneity of the AD space. In this sense, we call for more stringent AD analysis guidelines which require the incorporation of model error analysis schemes, to provide critical insight on the reliability of underlying AD algorithms. Additionally, any selected AD method should be rigorously validated to demonstrate its suitability for the model space over which it is applied. These steps will ultimately contribute to more accurate estimations of the reliability of model predictions. Finally, error analysis may also be useful in "rational" model refinement in that data expansion efforts and model retraining are focused on cohorts with the highest error rates.

The pursuit of accurate predictive models of the bioactivity of small molecules

Property prediction is a key interest in chemistry. For several decades there has been a continued and incremental development of mathematical models to predict properties. As more data is generated and accumulated, there seems to be more areas of opportunity to develop models with increased accuracy. The same is true if one considers the large developments in machine and deep learning models. However, along with the same areas of opportunity and development, issues and challenges remain and, with more data, new challenges emerge such as the quality and quantity and reliability of the data, and model reproducibility. Herein, we discuss the status of the accuracy of predictive models and present the authors' perspective of the direction of the field, emphasizing on good practices. We focus on predictive models of bioactive properties of small molecules relevant for drug discovery, agrochemical, food chemistry, natural product research, and related fields.

Aminopyrimidine Derivatives as Multiflavivirus Antiviral Compounds Identified from a Consensus Virtual Screening Approach

Around three billion people are at risk of infection by the dengue virus (DENV) and potentially other flaviviruses. Worldwide outbreaks of DENV, Zika virus (ZIKV), and yellow fever virus (YFV), the lack of antiviral drugs, and limitations on vaccine usage emphasize the need for novel antiviral research. Here, we propose a consensus virtual screening approach to discover potential protease inhibitors (NS3pro) against different flavivirus. We employed an in silico combination of a hologram quantitative structure-activity relationship (HQSAR) model and molecular docking on characterized binding sites followed by molecular dynamics (MD) simulations, which filtered a data set of 7.6 million compounds to 2,775 hits. Lastly, docking and MD simulations selected six final potential NS3pro inhibitors with stable interactions along the simulations. Five compounds had their antiviral activity confirmed against ZIKV, YFV, DENV-2, and DENV-3 (ranging from 4.21 ± 0.14 to 37.51 ± 0.8 μM), displaying aggregator characteristics for enzymatic inhibition against ZIKV NS3pro (ranging from 28 ± 7 to 70 ± 7 μM). Taken together, the compounds identified in this approach may contribute to the design of promising candidates to treat different flavivirus infections.

Development of predictive models for lymphedema by using blood tests and therapy data

Lymphedema is a disease that refers to tissue swelling caused by an accumulation of protein-rich fluid that is usually drained through the lymphatic system. Detection of lymphedema is often based on expensive diagnoses such as bioimpedance spectroscopy, shear wave elastography, computed tomography, etc. In current machine learning models for lymphedema prediction, reliance on observable symptoms reported by patients introduces the possibility of errors in patient-input data. Moreover, these symptoms are often absent during the initial stages of lymphedema, creating challenges in its early detection. Identifying lymphedema before these observable symptoms manifest would greatly benefit patients by potentially minimizing the discomfort caused by these symptoms. In this study, we propose to use new data, such as complete blood count, serum, and therapy data, to develop predictive models for lymphedema. This approach aims to compensate for the limitations of using only observable symptoms data. We collected data from 2137 patients, including 356 patients with lymphedema and 1781 patients without lymphedema, with the lymphedema status of each patient confirmed by clinicians. The data for each patient included: (1) a complete blood count (CBC) test, (2) a serum test, and (3) therapy information. We used various machine learning algorithms (i.e. random forest, gradient boosting, decision tree, logistic regression, and artificial neural network) to develop predictive models on the training dataset (i.e. 80% of the data) and evaluated the models on the external validation dataset (i.e. 20% of the data). After selecting the best predictive models, we created a web application to aid medical doctors and clinicians in the rapid screening of lymphedema patients. A dataset of 2137 patients was assembled from Seoul National University Bundang Hospital. Predictive models based on the random forest algorithm exhibited satisfactory performance (balanced accuracy = 87.0 ± 0.7%, sensitivity = 84.3 ± 0.6%, specificity = 89.1 ± 1.5%, precision = 97.4 ± 0.7%, F1 score = 90.4 ± 0.4%, and AUC = 0.931 ± 0.007). We developed a web application to facilitate the swift screening of lymphedema among medical practitioners: https://snubhtxt.shinyapps.io/SNUBH_Lymphedema . Our study introduces a novel tool for the early detection of lymphedema and establishes the foundation for future investigations into predicting different stages of the condition.

Designing molecules with autoencoder networks

Autoencoders are versatile tools in molecular informatics. These unsupervised neural networks serve diverse tasks such as data-driven molecular representation and constructive molecular design. This Review explores their algorithmic foundations and applications in drug discovery, highlighting the most active areas of development and the contributions autoencoder networks have made in advancing this field. We also explore the challenges and prospects concerning the utilization of autoencoders and the various adaptations of this neural network architecture in molecular design.

Finding Lead Compounds for Dengue Antivirals from a Collection of Old Drugs through In Silico Target Prediction and Subsequent In Vitro Validation

Dengue virus (DENV) infection is one of the most widely spread flavivirus infections. Despite the fatality it could cause, no antiviral treatment is currently available to treat the disease. Hence, this study aimed to repurpose old drugs as novel DENV NS3 inhibitors. Ligand-based (L-B) and proteochemometric (PCM) prediction models were built using 62,354 bioactivity data to screen for potential NS3 inhibitors. Selected drugs were then subjected to the foci forming unit reduction assay (FFURA) and protease inhibition assay. Finally, molecular docking was performed to validate these results. The in silico studies revealed that both models performed well in the internal and external validations. However, the L-B model showed better accuracy in the external validation in terms of its sensitivity (0.671). In the in vitro validation, all drugs (zileuton, trimethadione, and linalool) were able to moderately inhibit the viral activities at the highest concentration tested. Zileuton showed comparable results with linalool when tested at 2 mM against the DENV NS3 protease, with a reduction of protease activity at 17.89 and 18.42%, respectively. Two new compounds were also proposed through the combination of the selected drugs, which are ziltri (zilueton + trimethadione) and zilool (zileuton + linalool). The molecular docking study confirms the in vitro observations where all drugs and proposed compounds were able to achieve binding affinity ≥ -4.1 kcal/mol, with ziltri showing the highest affinity at -7.7 kcal/mol, surpassing the control, panduratin A. The occupation of both S1 and S2 subpockets of NS2B-NS3 may be essential and a reason for the lower binding energy shown by the proposed compounds compared to the screened drugs. Based on the results, this study provided five potential new lead compounds (ziltri, zilool, zileuton, linalool, and trimethadione) for DENV that could be modified further.

Classification-based machine learning approaches to predict the taste of molecules: A review

The capacity to discriminate safe from dangerous compounds has played an important role in the evolution of species, including human beings. Highly evolved senses such as taste receptors allow humans to navigate and survive in the environment through information that arrives to the brain through electrical pulses. Specifically, taste receptors provide multiple bits of information about the substances that are introduced orally. These substances could be pleasant or not according to the taste responses that they trigger. Tastes have been classified into basic (sweet, bitter, umami, sour and salty) or non-basic (astringent, chilling, cooling, heating, pungent), while some compounds are considered as multitastes, taste modifiers or tasteless. Classification-based machine learning approaches are useful tools to develop predictive mathematical relationships in such a way as to predict the taste class of new molecules based on their chemical structure. This work reviews the history of multicriteria quantitative structure-taste relationship modelling, starting from the first ligand-based (LB) classifier proposed in 1980 by Lemont B. Kier and concluding with the most recent studies published in 2022.

Analysis of structure-activity and structure-mechanism relationships among thyroid stimulating hormone receptor binding chemicals by leveraging the ToxCast library

The thyroid stimulating hormone receptor (TSHR) is crucial in thyroid hormone production in humans, and dysregulation in TSHR activation can lead to adverse health effects such as hypothyroidism and Graves' disease. Further, animal studies have shown that binding of endocrine disrupting chemicals (EDCs) with TSHR can lead to developmental toxicity. Hence, several such chemicals have been screened for their adverse physiological effects in human cell lines via high-throughput assays in the ToxCast project. The invaluable data generated by the ToxCast project has enabled the development of toxicity predictors, but they can be limited in their predictive ability due to the heterogeneity in structure-activity relationships among chemicals. Here, we systematically investigated the heterogeneity in structure-activity as well as structure-mechanism relationships among the TSHR binding chemicals from ToxCast. By employing a structure-activity similarity (SAS) map, we identified 79 activity cliffs among 509 chemicals in TSHR agonist dataset and 69 activity cliffs among 650 chemicals in the TSHR antagonist dataset. Further, by using the matched molecular pair (MMP) approach, we find that the resultant activity cliffs (MMP-cliffs) are a subset of activity cliffs identified via the SAS map approach. Subsequently, by leveraging ToxCast mechanism of action (MOA) annotations for chemicals common to both TSHR agonist and TSHR antagonist datasets, we identified 3 chemical pairs as strong MOA-cliffs and 19 chemical pairs as weak MOA-cliffs. In conclusion, the insights from this systematic investigation of the TSHR binding chemicals are likely to inform ongoing efforts towards development of better predictive toxicity models for characterization of the chemical exposome.

Dynamic applicability domain (dAD): compound-target binding affinity estimates with local conformal prediction

MOTIVATION

Increasing efforts are being made in the field of machine learning to advance the learning of robust and accurate models from experimentally measured data and enable more efficient drug discovery processes. The prediction of binding affinity is one of the most frequent tasks of compound bioactivity modelling. Learned models for binding affinity prediction are assessed by their average performance on unseen samples, but point predictions are typically not provided with a rigorous confidence assessment. Approaches, such as the conformal predictor framework equip conventional models with a more rigorous assessment of confidence for individual point predictions. In this article, we extend the inductive conformal prediction framework for interaction data, in particular the compound-target binding affinity prediction task. The new framework is based on dynamically defined calibration sets that are specific for each testing pair and provides prediction assessment in the context of calibration pairs from its compound-target neighbourhood, enabling improved estimates based on the local properties of the prediction model.

RESULTS

The effectiveness of the approach is benchmarked on several publicly available datasets and tested in realistic use-case scenarios with increasing levels of difficulty on a complex compound-target binding affinity space. We demonstrate that in such scenarios, novel approach combining applicability domain paradigm with conformal prediction framework, produces superior confidence assessment with valid and more informative prediction regions compared to other 'state-of-the-art' conformal prediction approaches.

AVAILABILITY AND IMPLEMENTATION

Dataset and the code are available on GitHub (https://github.com/mlkr-rbi/dAD).

New Insights on the Activity and Selectivity of MAO-B Inhibitors through In Silico Methods

To facilitate the identification of novel MAO-B inhibitors, we elaborated a consolidated computational approach, including a pharmacophoric atom-based 3D quantitative structure-activity relationship (QSAR) model, activity cliffs, fingerprint, and molecular docking analysis on a dataset of 126 molecules. An AAHR.2 hypothesis with two hydrogen bond acceptors (A), one hydrophobic (H), and one aromatic ring (R) supplied a statistically significant 3D QSAR model reflected by the parameters: R² = 0.900 (training set); Q² = 0.774 and Pearson's R = 0.884 (test set), stability s = 0.736. Hydrophobic and electron-withdrawing fields portrayed the relationships between structural characteristics and inhibitory activity. The quinolin-2-one scaffold has a key role in selectivity towards MAO-B with an AUC of 0.962, as retrieved by ECFP4 analysis. Two activity cliffs showing meaningful potency variation in the MAO-B chemical space were observed. The docking study revealed interactions with crucial residues TYR:435, TYR:326, CYS:172, and GLN:206 responsible for MAO-B activity. Molecular docking is in consensus with and complementary to pharmacophoric 3D QSAR, ECFP4, and MM-GBSA analysis. The computational scenario provided here will assist chemists in quickly designing and predicting new potent and selective candidates as MAO-B inhibitors for MAO-B-driven diseases. This approach can also be used to identify MAO-B inhibitors from other libraries or screen top molecules for other targets involved in suitable diseases.

Machine Learning Model for Screening Thyroid Stimulating Hormone Receptor Agonists Based on Updated Datasets and Improved Applicability Domain Metrics

Machine learning (ML) models for screening endocrine-disrupting chemicals (EDCs), such as thyroid stimulating hormone receptor (TSHR) agonists, are essential for sound management of chemicals. Previous models for screening TSHR agonists were built on imbalanced datasets and lacked applicability domain (AD) characterization essential for regulatory application. Herein, an updated TSHR agonist dataset was built, for which the ratio of active to inactive compounds greatly increased to 1:2.6, and chemical spaces of structure-activity landscapes (SALs) were enhanced. Resulting models based on 7 molecular representations and 4 ML algorithms were proven to outperform previous ones. Weighted similarity density (ρ_s) and weighted inconsistency of activities (I_A) were proposed to characterize the SALs, and a state-of-the-art AD characterization methodology AD_SAL{ρ_s, I_A} was established. An optimal classifier developed with PubChem fingerprints and the random forest algorithm, coupled with AD_SAL{ρ_s ≥ 0.15, I_A ≤ 0.65}, exhibited good performance on the validation set with the area under the receiver operating characteristic curve being 0.984 and balanced accuracy being 0.941 and identified 90 TSHR agonist classes that could not be found previously. The classifier together with the AD_SAL{ρ_s, I_A} may serve as efficient tools for screening EDCs, and the AD characterization methodology may be applied to other ML models.

Large-scale evaluation of k-fold cross-validation ensembles for uncertainty estimation

It is insightful to report an estimator that describes how certain a model is in a prediction, additionally to the prediction alone. For regression tasks, most approaches implement a variation of the ensemble method, apart from few exceptions. Instead of a single estimator, a group of estimators yields several predictions for an input. The uncertainty can then be quantified by measuring the disagreement between the predictions, for example by the standard deviation. In theory, ensembles should not only provide uncertainties, they also boost the predictive performance by reducing errors arising from variance. Despite the development of novel methods, they are still considered the "golden-standard" to quantify the uncertainty of regression models. Subsampling-based methods to obtain ensembles can be applied to all models, regardless whether they are related to deep learning or traditional machine learning. However, little attention has been given to the question whether the ensemble method is applicable to virtually all scenarios occurring in the field of cheminformatics. In a widespread and diversified attempt, ensembles are evaluated for 32 datasets of different sizes and modeling difficulty, ranging from physicochemical properties to biological activities. For increasing ensemble sizes with up to 200 members, the predictive performance as well as the applicability as uncertainty estimator are shown for all combinations of five modeling techniques and four molecular featurizations. Useful recommendations were derived for practitioners regarding the success and minimum size of ensembles, depending on whether predictive performance or uncertainty quantification is of more importance for the task at hand.

Predictive Modeling of PROTAC Cell Permeability with Machine Learning

Approaches for predicting proteolysis targeting chimera (PROTAC) cell permeability are of major interest to reduce resource-demanding synthesis and testing of low-permeable PROTACs. We report a comprehensive investigation of the scope and limitations of machine learning-based binary classification models developed using 17 simple descriptors for large and structurally diverse sets of cereblon (CRBN) and von Hippel-Lindau (VHL) PROTACs. For the VHL PROTAC set, kappa nearest neighbor and random forest models performed best and predicted the permeability of a blinded test set with >80% accuracy (k ≥ 0.57). Models retrained by combining the original training and the blinded test set performed equally well for a second blinded VHL set. However, models for CRBN PROTACs were less successful, mainly due to the imbalanced nature of the CRBN datasets. All descriptors contributed to the models, but size and lipophilicity were the most important. We conclude that properly trained machine learning models can be integrated as effective filters in the PROTAC design process.

Pulled fly balls are harder to catch: a game analysis with a machine learning approach

Two hypotheses were tested: (1) the deflecting motion of fly balls caused by aerodynamic effects varies between the pull side and opposite side of the fair territory, and (2) the probability of flyout is lower on the pull side than the opposite side in Japan’s professional baseball games. From all radar-tracking outputs of official games in 2018–2019, fly balls that resulted in outs or base hits were selected for analysis (N = 25,413), and indices representing horizontal and vertical deflecting motions of fly balls were computed and compared between the pull side and opposite side. A machine learning algorithm was used to construct a model to predict the probability of flyout from the kinematic characteristics of fly balls. Flyout zones where the probability of flyout was > 0.6 were computed for a systematically constructed set of fly balls having identical distribution between the pull side and opposite side. The results showed that: (1) most fly balls landing on the opposite side deflected in the same direction whereas the pulled fly balls deflected to either direction, (2) the pulled low fly balls had greater variability in the deflecting motions than the opposite side counterpart, (3) overall probability of flyout of the low fly balls was lower in the pull side (0.41) than the opposite side (0.49), and (4) the flyout zone of an outfielder in the pull side (mean = 698 m2) for low fly balls was smaller than that of the others (≥ 779 m2). The hypotheses were supported. The pulled low fly balls had substantial variations in the direction and magnitude of deflections, which might have reduced the flyout zone on the pull side.

Predicting the Bioconcentration Factor in Fish from Molecular Structures

The bioconcentration factor (BCF) is one of the metrics used to evaluate the potential of a substance to bioaccumulate into aquatic organisms. In this work, linear and non-linear regression QSARs were developed for the prediction of log BCF using different computational approaches, and starting from a large and structurally heterogeneous dataset. The new MLR-OLS and ANN regression models have good fitting with R2 values of 0.62 and 0.70, respectively, and comparable external predictivity with R2ext 0.64 and 0.65 (RMSEext of 0.78 and 0.76), respectively. Furthermore, linear and non-linear classification models were developed using the regulatory threshold BCF >2000. A class balanced subset was used to develop classification models which were applied to chemicals not used to create the QSARs. These classification models are characterized by external and internal accuracy up to 84% and 90%, respectively, and sensitivity and specificity up to 90% and 80%, respectively. QSARs presented in this work are validated according to regulatory requirements and their quality is in line with other tools available for the same endpoint and dataset, with the advantage of low complexity and easy application through the software QSAR-ME Profiler. These QSARs can be used as alternatives for, or in combination with, existing models to support bioaccumulation assessment procedures.

Investigation of a Data Split Strategy Involving the Time Axis in Adverse Event Prediction Using Machine Learning

Adverse events are a serious issue in drug development, and many prediction methods using machine learning have been developed. The random split cross-validation is the de facto standard for model building and evaluation in machine learning, but care should be taken in adverse event prediction because this approach does not strictly match the real-world situation. The time split, which uses the time axis, is considered suitable for real-world prediction. However, the differences in model performance obtained using the time and random splits are not clear due to the lack of comparable studies. To understand the differences, we compared the model performance between the time and random splits using nine types of compound information as input, eight adverse events as targets, and six machine learning algorithms. The random split showed higher area under the curve values than did the time split for six of eight targets. The chemical spaces of the training and test datasets of the time split were similar, suggesting that the concept of applicability domain is insufficient to explain the differences derived from the splitting. The area under the curve differences were smaller for the protein interaction than for the other datasets. Subsequent detailed analyses suggested the danger of confounding in the use of knowledge-based information in the time split. These findings indicate the importance of understanding the differences between the time and random splits in adverse event prediction and suggest that appropriate use of the splitting strategies and interpretation of results are necessary for the real-world prediction of adverse events. We provide the analysis code and datasets used in the present study at https://github.com/mizuno-group/AE_prediction.

Studying and mitigating the effects of data drifts on ML model performance at the example of chemical toxicity data

Machine learning models are widely applied to predict molecular properties or the biological activity of small molecules on a specific protein. Models can be integrated in a conformal prediction (CP) framework which adds a calibration step to estimate the confidence of the predictions. CP models present the advantage of ensuring a predefined error rate under the assumption that test and calibration set are exchangeable. In cases where the test data have drifted away from the descriptor space of the training data, or where assay setups have changed, this assumption might not be fulfilled and the models are not guaranteed to be valid. In this study, the performance of internally valid CP models when applied to either newer time-split data or to external data was evaluated. In detail, temporal data drifts were analysed based on twelve datasets from the ChEMBL database. In addition, discrepancies between models trained on publicly-available data and applied to proprietary data for the liver toxicity and MNT in vivo endpoints were investigated. In most cases, a drastic decrease in the validity of the models was observed when applied to the time-split or external (holdout) test sets. To overcome the decrease in model validity, a strategy for updating the calibration set with data more similar to the holdout set was investigated. Updating the calibration set generally improved the validity, restoring it completely to its expected value in many cases. The restored validity is the first requisite for applying the CP models with confidence. However, the increased validity comes at the cost of a decrease in model efficiency, as more predictions are identified as inconclusive. This study presents a strategy to recalibrate CP models to mitigate the effects of data drifts. Updating the calibration sets without having to retrain the model has proven to be a useful approach to restore the validity of most models.

Development of Predictive Classification Models for Whole Cell Antimycobacterial Activity of Benzothiazinones

Nitrobenzothiazinones (BTZs) are a very potent class of antibiotics against Mycobacterium tuberculosis. However, relationships between their structural properties and whole cell activity remain poorly predictable. Herein, we present the synthesis and antimycobacterial evaluation of a diverse set of BTZs. High potency was predominantly achieved by piperidine and piperazine substitutions, whereupon three compounds were identified as promising candidates, showing preferable metabolic stability. Lack of correlation between potency and calculated binding energies suggested that target inhibition is not the only requirement to obtain suitable antimycobacterial agents. In contrast, prediction of whole cell activity class was successfully accomplished by extensively validated machine learning models. The performance of the superior model was further verified by >70% correct class predictions for a large set of reported BTZs. Our generated model is thus a key prerequisite to streamline lead optimization endeavors, particularly regarding the improvement of overall hit rates in whole cell antimycobacterial assays.

Semi-automated workflow for molecular pair analysis and QSAR-assisted transformation space expansion

In the process of drug discovery, the optimization of lead compounds has always been a challenge faced by pharmaceutical chemists. Matched molecular pair analysis (MMPA), a promising tool to efficiently extract and summarize the relationship between structural transformation and property change, is suitable for local structural optimization tasks. Especially, the integration of MMPA with QSAR modeling can further strengthen the utility of MMPA in molecular optimization navigation. In this study, a new semi-automated procedure based on KNIME was developed to support MMPA on both large- and small-scale datasets, including molecular preparation, QSAR model construction, applicability domain evaluation, and MMP calculation and application. Two examples covering regression and classification tasks were provided to gain a better understanding of the importance of MMPA, which has also shown the reliability and utility of this MMPA-by-QSAR pipeline.

Validation strategies for target prediction methods

Computational methods for target prediction, based on molecular similarity and network-based approaches, machine learning, docking and others, have evolved as valuable and powerful tools to aid the challenging task of mode of action identification for bioactive small molecules such as drugs and drug-like compounds. Critical to discerning the scope and limitations of a target prediction method is understanding how its performance was evaluated and reported. Ideally, large-scale prospective experiments are conducted to validate the performance of a model; however, this expensive and time-consuming endeavor is often not feasible. Therefore, to estimate the predictive power of a method, statistical validation based on retrospective knowledge is commonly used. There are multiple statistical validation techniques that vary in rigor. In this review we discuss the validation strategies employed, highlighting the usefulness and constraints of the validation schemes and metrics that are employed to measure and describe performance. We address the limitations of measuring only generalized performance, given that the underlying bioactivity and structural data are biased towards certain small-molecule scaffolds and target families, and suggest additional aspects of performance to consider in order to produce more detailed and realistic estimates of predictive power. Finally, we describe the validation strategies that were employed by some of the most thoroughly validated and accessible target prediction methods.

In Silico Prediction of CYP2C8 Inhibition with Machine-Learning Methods

Cytochrome P450 2C8 (CYP2C8) is a major drug-metabolizing enzyme in humans and is responsible for the metabolism of ∼5% drugs in clinical use. Thus, inhibition of CYP2C8, which causes potential adverse drug events, cannot be neglected. The in vitro drug interaction studies guidelines for industry issued by the FDA also point out that it needs to be determined whether investigated drugs are CYP2C8 inhibitors before clinical trials. However, current studies mainly focus on predicting the inhibitors of other major P450 enzymes, and the importance of CYP2C8 inhibition has been overlooked. Therefore, there is a need to develop models for identifying potential CYP2C8 inhibition. In this study, in silico classification models for predicting CYP2C8 inhibition were built by five machine-learning methods combined with nine molecular fingerprints. The performance of the models built was evaluated by test and external validation sets. The best model had AUC values of 0.85 and 0.90 for the test and external validation sets, respectively. The applicability domain was analyzed based on the molecular similarity and exhibited an impact on the improvement of prediction accuracy. Furthermore, several representative privileged substructures such as 1H-benzo[d]imidazole, 1-phenyl-1H-pyrazole, and quinoline were identified by information gain and substructure frequency analysis. Overall, our results would be helpful for the prediction of CYP2C8 inhibition.

Schistosomiasis Drug Discovery in the Era of Automation and Artificial Intelligence

Schistosomiasis is a parasitic disease caused by trematode worms of the genus Schistosoma and affects over 200 million people worldwide. The control and treatment of this neglected tropical disease is based on a single drug, praziquantel, which raises concerns about the development of drug resistance. This, and the lack of efficacy of praziquantel against juvenile worms, highlights the urgency for new antischistosomal therapies. In this review we focus on innovative approaches to the identification of antischistosomal drug candidates, including the use of automated assays, fragment-based screening, computer-aided and artificial intelligence-based computational methods. We highlight the current developments that may contribute to optimizing research outputs and lead to more effective drugs for this highly prevalent disease, in a more cost-effective drug discovery endeavor.

Assessing the calibration in toxicological in vitro models with conformal prediction

Machine learning methods are widely used in drug discovery and toxicity prediction. While showing overall good performance in cross-validation studies, their predictive power (often) drops in cases where the query samples have drifted from the training data’s descriptor space. Thus, the assumption for applying machine learning algorithms, that training and test data stem from the same distribution, might not always be fulfilled. In this work, conformal prediction is used to assess the calibration of the models. Deviations from the expected error may indicate that training and test data originate from different distributions. Exemplified on the Tox21 datasets, composed of chronologically released Tox21Train, Tox21Test and Tox21Score subsets, we observed that while internally valid models could be trained using cross-validation on Tox21Train, predictions on the external Tox21Score data resulted in higher error rates than expected. To improve the prediction on the external sets, a strategy exchanging the calibration set with more recent data, such as Tox21Test, has successfully been introduced. We conclude that conformal prediction can be used to diagnose data drifts and other issues related to model calibration. The proposed improvement strategy—exchanging the calibration data only—is convenient as it does not require retraining of the underlying model.

QSAR-Co-X: an open source toolkit for multitarget QSAR modelling

Quantitative structure activity relationships (QSAR) modelling is a well-known computational tool, often used in a wide variety of applications. Yet one of the major drawbacks of conventional QSAR modelling is that models are set up based on a limited number of experimental and/or theoretical conditions. To overcome this, the so-called multitasking or multitarget QSAR (mt-QSAR) approaches have emerged as new computational tools able to integrate diverse chemical and biological data into a single model equation, thus extending and improving the reliability of this type of modelling. We have developed QSAR-Co-X, an open source python–based toolkit (available to download at https://github.com/ncordeirfcup/QSAR-Co-X) for supporting mt-QSAR modelling following the Box-Jenkins moving average approach. The new toolkit embodies several functionalities for dataset selection and curation plus computation of descriptors, for setting up linear and non-linear models, as well as for a comprehensive results analysis. The workflow within this toolkit is guided by a cohort of multiple statistical parameters and graphical outputs onwards assessing both the predictivity and the robustness of the derived mt-QSAR models. To monitor and demonstrate the functionalities of the designed toolkit, four case-studies pertaining to previously reported datasets are examined here. We believe that this new toolkit, along with our previously launched QSAR-Co code, will significantly contribute to make mt-QSAR modelling widely and routinely applicable.

MetaTREE, a Novel Database Focused on Metabolic Trees, Predicts an Important Detoxification Mechanism: The Glutathione Conjugation

(1) Background: Data accuracy plays a key role in determining the model performances and the field of metabolism prediction suffers from the lack of truly reliable data. To enhance the accuracy of metabolic data, we recently proposed a manually curated database collected by a meta-analysis of the specialized literature (MetaQSAR). Here we aim to further increase data accuracy by focusing on publications reporting exhaustive metabolic trees. This selection should indeed reduce the number of false negative data. (2) Methods: A new metabolic database (MetaTREE) was thus collected and utilized to extract a dataset for metabolic data concerning glutathione conjugation (MT-dataset). After proper pre-processing, this dataset, along with the corresponding dataset extracted from MetaQSAR (MQ-dataset), was utilized to develop binary classification models using a random forest algorithm. (3) Results: The comparison of the models generated by the two collected datasets reveals the better performances reached by the MT-dataset (MCC raised from 0.63 to 0.67, sensitivity from 0.56 to 0.58). The analysis of the applicability domain also confirms that the model based on the MT-dataset shows a more robust predictive power with a larger applicability domain. (4) Conclusions: These results confirm that focusing on metabolic trees represents a convenient approach to increase data accuracy by reducing the false negative cases. The encouraging performances shown by the models developed by the MT-dataset invites to use of MetaTREE for predictive studies in the field of xenobiotic metabolism.

The challenges of generalizability in artificial intelligence for ADME/Tox endpoint and activity prediction

INTRODUCTION

Artificial intelligence (AI) has seen a massive resurgence in recent years with wide successes in computer vision, natural language processing, and games. The similar creation of robust and accurate AI models for ADME/Tox endpoint and activity prediction would be revolutionary to drug discovery pipelines. There have been numerous demonstrations of successful applications, but a key challenge remains: how generalizable are these predictive models?

AREAS COVERED

The authors present a summary of current promising components of AI models in the context of early drug discovery where ADME/Tox endpoint and activity prediction is the main driver of the iterative drug design process. Following that is a review of applicability domains and dataset construction considerations which determine generalizability bottlenecks for AI deployment. Further reviewed is the role of promising learning frameworks - multitask, transfer, and meta learning - which leverage auxiliary data to overcome issues of generalizability.

EXPERT OPINION

The authors conclude that the most promising direction toward integrating reliable and informative AI models into the drug discovery pipeline is a conjunction of learned feature representations, deep learning, and novel learning frameworks. Such a solution would address the sparse and incomplete datasets that are available for key endpoints related to drug discovery.

Artificial Intelligence Applied to the Rapid Identification of New Antimalarial Candidates with Dual-Stage Activity

Increasing reports of multidrug-resistant malaria parasites urge the discovery of new effective drugs with different chemical scaffolds. Protein kinases play a key role in many cellular processes such as signal transduction and cell division, making them interesting targets in many diseases. Protein kinase 7 (PK7) is an orphan kinase from the Plasmodium genus, essential for the sporogonic cycle of these parasites. Here, we applied a robust and integrative artificial intelligence-assisted virtual-screening (VS) approach using shape-based and machine learning models to identify new potential PK7 inhibitors with in vitro antiplasmodial activity. Eight virtual hits were experimentally evaluated, and compound LabMol-167 inhibited ookinete conversion of Plasmodium berghei and blood stages of Plasmodium falciparum at nanomolar concentrations with low cytotoxicity in mammalian cells. As PK7 does not have an essential role in the Plasmodium blood stage and our virtual screening strategy aimed for both PK7 and blood-stage inhibition, we conducted an in silico target fishing approach and propose that this compound might also inhibit P. falciparum PK5, acting as a possible dual-target inhibitor. Finally, docking studies of LabMol-167 with P. falciparum PK7 and PK5 proteins highlighted key interactions for further hit-to lead optimization.

Novel Development of Predictive Feature Fingerprints to Identify Chemistry-Based Features for the Effective Drug Design of SARS-CoV-2 Target Antagonists and Inhibitors Using Machine Learning

A unique approach to bioactivity and chemical data curation coupled with random forest analyses has led to a series of target-specific and cross-validated predictive feature fingerprints (PFF) that have high predictability across multiple therapeutic targets and disease stages involved in the severe acute respiratory syndrome due to coronavirus 2 (SARS-CoV-2)-induced COVID-19 pandemic, which include plasma kallikrein, human immunodeficiency virus (HIV)-protease, nonstructural protein (NSP)5, NSP12, Janus kinase (JAK) family, and AT-1. The approach was highly accurate in determining the matched target for the different compound sets and suggests that the models could be used for virtual screening of target-specific compound libraries. The curation-modeling process was successfully applied to a SARS-CoV-2 phenotypic screen and could be used for predictive bioactivity estimation and prioritization for clinical trial selection; virtual screening of drug libraries for the repurposing of drug molecules; and analysis and direction of proprietary data sets.

Machine learning meets mechanistic modelling for accurate prediction of experimental activation energies

Accurate prediction of chemical reactions in solution is challenging for current state-of-the-art approaches based on transition state modelling with density functional theory. Models based on machine learning have emerged as a promising alternative to address these problems, but these models currently lack the precision to give crucial information on the magnitude of barrier heights, influence of solvents and catalysts and extent of regio- and chemoselectivity. Here, we construct hybrid models which combine the traditional transition state modelling and machine learning to accurately predict reaction barriers. We train a Gaussian Process Regression model to reproduce high-quality experimental kinetic data for the nucleophilic aromatic substitution reaction and use it to predict barriers with a mean absolute error of 0.77 kcal mol-1 for an external test set. The model was further validated on regio- and chemoselectivity prediction on patent reaction data and achieved a competitive top-1 accuracy of 86%, despite not being trained explicitly for this task. Importantly, the model gives error bars for its predictions that can be used for risk assessment by the end user. Hybrid models emerge as the preferred alternative for accurate reaction prediction in the very common low-data situation where only 100-150 rate constants are available for a reaction class. With recent advances in deep learning for quickly predicting barriers and transition state geometries from density functional theory, we envision that hybrid models will soon become a standard alternative to complement current machine learning approaches based on ground-state physical organic descriptors or structural information such as molecular graphs or fingerprints.

In silico prediction of mitochondrial toxicity of chemicals using machine learning methods

Mitochondria are important organelles in human cells, providing more than 95% of the energy. However, some drugs and environmental chemicals could induce mitochondrial dysfunction, which might cause complex diseases and even worsen the condition of patients with mitochondrial damage. Some drugs have been withdrawn from the market due to their severe mitochondrial toxicity, such as troglitazone. Therefore, there is an urgent need to develop models that could accurately predict the mitochondrial toxicity of chemicals. In this paper, suitable data were obtained from literature and databases first. Then nine types of fingerprints were used to characterize these compounds. Finally, different algorithms were used to build models. Meanwhile, the applicability domain of the prediction models was defined. We have also explored the structural alerts of mitochondrial toxicity, which would be helpful for medicinal chemists to better predict mitochondrial toxicity and further optimize lead compounds.

In Silico Approach to Predict Severe Cutaneous Adverse Reactions Using the Japanese Adverse Drug Event Report Database

Severe cutaneous adverse reactions (SCARs), such as Stevens–Johnson syndrome/toxic epidermal necrolysis and drug‐induced hypersensitivity syndrome, are rare and occasionally fatal. However, it is difficult to detect SCARs at the drug development stage, necessitating a new approach for prediction. Therefore, in this study, using the chemical structure information of SCAR‐causative drugs from the Japanese Adverse Drug Event Report (JADER) database, we tried to develop a predictive classification model of SCAR through deep learning. In the JADER database from 2004 to 2017, we defined 185 SCAR‐positive drugs and 195 SCAR‐negative drugs using proportional reporting ratios as the signal detection method, and the total number of reports. These SCAR‐positive and SCAR‐negative drugs were randomly divided into the training dataset for model construction and the test dataset for evaluation. The model performance was evaluated in the independent test dataset inside the applicability domain (AD), which is the chemical space for reliable prediction results. Using the deep learning model with molecular descriptors as the drug structure information, the area under the curve was 0.76 for the 148 drugs of the test dataset inside the AD. The method developed in the present study allows for utilizing the JADER database for SCAR classification, with potential to improve screening efficiency in the development of new drugs. This method may also help to noninvasively identify the causative drug, and help assess the causality between drugs and SCARs in postmarketing surveillance.

Comprehensive Analysis of Applicability Domains of QSPR Models for Chemical Reactions

Nowadays, the problem of the model's applicability domain (AD) definition is an active research topic in chemoinformatics. Although many various AD definitions for the models predicting properties of molecules (Quantitative Structure-Activity/Property Relationship (QSAR/QSPR) models) were described in the literature, no one for chemical reactions (Quantitative Reaction-Property Relationships (QRPR)) has been reported to date. The point is that a chemical reaction is a much more complex object than an individual molecule, and its yield, thermodynamic and kinetic characteristics depend not only on the structures of reactants and products but also on experimental conditions. The QRPR models' performance largely depends on the way that chemical transformation is encoded. In this study, various AD definition methods extensively used in QSAR/QSPR studies of individual molecules, as well as several novel approaches suggested in this work for reactions, were benchmarked on several reaction datasets. The ability to exclude wrong reaction types, increase coverage, improve the model performance and detect Y-outliers were tested. As a result, several "best" AD definitions for the QRPR models predicting reaction characteristics have been revealed and tested on a previously published external dataset with a clear AD definition problem.

Discriminant models on mitochondrial toxicity improved by consensus modeling and resolving imbalance in training

Humans and animals may be exposed to tens of thousands of natural and synthetic chemicals during their lifespan. It is difficult to assess risk for all the chemicals with experimental toxicity tests. An alternative approach is to use computational toxicology methods such as quantitative structure-activity relationship (QSAR) modeling. Mitochondrial toxicity is involved in many diseases such as cancer, neurodegeneration, type 2 diabetes, cardiovascular diseases and autoimmune diseases. Thus, it is important to rapidly and efficiently identify chemicals with mitochondrial toxicity. In this study, five machine learning algorithms and twelve types of molecular fingerprints were employed to generate QSAR discriminant models for mitochondrial toxicity. A threshold moving method was adopted to resolve the imbalance issue in the training data. Consensus of the models by an averaging probability strategy improved prediction performance. The best model has correct classification rates of 81.8% and 88.3% in ten-fold cross validation and external validation, respectively. Substructures such as phenol, carboxylic acid, nitro and arylchloride were found informative through analysis of information gain and frequency of substructures. The results demonstrate that resolving imbalance in training and building consensus models can improve classification rates for mitochondrial toxicity prediction.