Advances, opportunities, and challenges in methods for interrogating the structure activity relationships of natural products

Time span in literature: 1985-early 2024Natural products play a key role in drug discovery, both as a direct source of drugs and as a starting point for the development of synthetic compounds. Most natural products are not suitable to be used as drugs without further modification due to insufficient activity or poor pharmacokinetic properties. Choosing what modifications to make requires an understanding of the compound's structure-activity relationships. Use of structure-activity relationships is commonplace and essential in medicinal chemistry campaigns applied to human-designed synthetic compounds. Structure-activity relationships have also been used to improve the properties of natural products, but several challenges still limit these efforts. Here, we review methods for studying the structure-activity relationships of natural products and their limitations. Specifically, we will discuss how synthesis, including total synthesis, late-stage derivatization, chemoenzymatic synthetic pathways, and engineering and genome mining of biosynthetic pathways can be used to produce natural product analogs and discuss the challenges of each of these approaches. Finally, we will discuss computational methods including machine learning methods for analyzing the relationship between biosynthetic genes and product activity, computer aided drug design techniques, and interpretable artificial intelligence approaches towards elucidating structure-activity relationships from models trained to predict bioactivity from chemical structure. Our focus will be on these latter topics as their applications for natural products have not been extensively reviewed. We suggest that these methods are all complementary to each other, and that only collaborative efforts using a combination of these techniques will result in a full understanding of the structure-activity relationships of natural products.

Development and validation of AI/ML derived splice-switching oligonucleotides

Splice-switching oligonucleotides (SSOs) are antisense compounds that act directly on pre-mRNA to modulate alternative splicing (AS). This study demonstrates the value that artificial intelligence/machine learning (AI/ML) provides for the identification of functional, verifiable, and therapeutic SSOs. We trained XGboost tree models using splicing factor (SF) pre-mRNA binding profiles and spliceosome assembly information to identify modulatory SSO binding sites on pre-mRNA. Using Shapley and out-of-bag analyses we also predicted the identity of specific SFs whose binding to pre-mRNA is blocked by SSOs. This step adds considerable transparency to AI/ML-driven drug discovery and informs biological insights useful in further validation steps. We applied this approach to previously established functional SSOs to retrospectively identify the SFs likely to regulate those events. We then took a prospective validation approach using a novel target in triple negative breast cancer (TNBC), NEDD4L exon 13 (NEDD4Le13). Targeting NEDD4Le13 with an AI/ML-designed SSO decreased the proliferative and migratory behavior of TNBC cells via downregulation of the TGFβ pathway. Overall, this study illustrates the ability of AI/ML to extract actionable insights from RNA-seq data.

Interpreting drug synergy in breast cancer with deep learning using target-protein inhibition profiles

BACKGROUND

Breast cancer is the most common malignancy among women worldwide. Despite advances in treating breast cancer over the past decades, drug resistance and adverse effects remain challenging. Recent therapeutic progress has shifted toward using drug combinations for better treatment efficiency. However, with a growing number of potential small-molecule cancer inhibitors, in silico strategies to predict pharmacological synergy before experimental trials are required to compensate for time and cost restrictions. Many deep learning models have been previously proposed to predict the synergistic effects of drug combinations with high performance. However, these models heavily relied on a large number of drug chemical structural fingerprints as their main features, which made model interpretation a challenge.

RESULTS

This study developed a deep neural network model that predicts synergy between small-molecule pairs based on their inhibitory activities against 13 selected key proteins. The synergy prediction model achieved a Pearson correlation coefficient between model predictions and experimental data of 0.63 across five breast cancer cell lines. BT-549 and MCF-7 achieved the highest correlation of 0.67 when considering individual cell lines. Despite achieving a moderate correlation compared to previous deep learning models, our model offers a distinctive advantage in terms of interpretability. Using the inhibitory activities against key protein targets as the main features allowed a straightforward interpretation of the model since the individual features had direct biological meaning. By tracing the synergistic interactions of compounds through their target proteins, we gained insights into the patterns our model recognized as indicative of synergistic effects.

CONCLUSIONS

The framework employed in the present study lays the groundwork for future advancements, especially in model interpretation. By combining deep learning techniques and target-specific models, this study shed light on potential patterns of target-protein inhibition profiles that could be exploited in breast cancer treatment.

Finding the best trade-off between performance and interpretability in predicting hospital length of stay using structured and unstructured data

OBJECTIVE

This study aims to develop high-performing Machine Learning and Deep Learning models in predicting hospital length of stay (LOS) while enhancing interpretability. We compare performance and interpretability of models trained only on structured tabular data with models trained only on unstructured clinical text data, and on mixed data.

METHODS

The structured data was used to train fourteen classical Machine Learning models including advanced ensemble trees, neural networks and k-nearest neighbors. The unstructured data was used to fine-tune a pre-trained Bio Clinical BERT Transformer Deep Learning model. The structured and unstructured data were then merged into a tabular dataset after vectorization of the clinical text and a dimensional reduction through Latent Dirichlet Allocation. The study used the free and publicly available Medical Information Mart for Intensive Care (MIMIC) III database, on the open AutoML Library AutoGluon. Performance is evaluated with respect to two types of random classifiers, used as baselines.

RESULTS

The best model from structured data demonstrates high performance (ROC AUC = 0.944, PRC AUC = 0.655) with limited interpretability, where the most important predictors of prolonged LOS are the level of blood urea nitrogen and of platelets. The Transformer model displays a good but lower performance (ROC AUC = 0.842, PRC AUC = 0.375) with a richer array of interpretability by providing more specific in-hospital factors including procedures, conditions, and medical history. The best model trained on mixed data satisfies both a high level of performance (ROC AUC = 0.963, PRC AUC = 0.746) and a much larger scope in interpretability including pathologies of the intestine, the colon, and the blood; infectious diseases, respiratory problems, procedures involving sedation and intubation, and vascular surgery.

CONCLUSIONS

Our results outperform most of the state-of-the-art models in LOS prediction both in terms of performance and of interpretability. Data fusion between structured and unstructured text data may significantly improve performance and interpretability.

Explaining compound activity predictions with a substructure-aware loss for graph neural networks

Explainable machine learning is increasingly used in drug discovery to help rationalize compound property predictions. Feature attribution techniques are popular choices to identify which molecular substructures are responsible for a predicted property change. However, established molecular feature attribution methods have so far displayed low performance for popular deep learning algorithms such as graph neural networks (GNNs), especially when compared with simpler modeling alternatives such as random forests coupled with atom masking. To mitigate this problem, a modification of the regression objective for GNNs is proposed to specifically account for common core structures between pairs of molecules. The presented approach shows higher accuracy on a recently-proposed explainability benchmark. This methodology has the potential to assist with model explainability in drug discovery pipelines, particularly in lead optimization efforts where specific chemical series are investigated.

Explainable ML models for a deeper insight on treatment decision for localized prostate cancer

Although there are several decision aids for the treatment of localized prostate cancer (PCa), there are limitations in the consistency and certainty of the information provided. We aimed to better understand the treatment decision process and develop a decision-predicting model considering oncologic, demographic, socioeconomic, and geographic factors. Men newly diagnosed with localized PCa between 2010 and 2015 from the Surveillance, Epidemiology, and End Results Prostate with Watchful Waiting database were included (n = 255,837). We designed two prediction models: (1) Active surveillance/watchful waiting (AS/WW), radical prostatectomy (RP), and radiation therapy (RT) decision prediction in the entire cohort. (2) Prediction of AS/WW decisions in the low-risk cohort. The discrimination of the model was evaluated using the multiclass area under the curve (AUC). A plausible Shapley additive explanations value was used to explain the model's prediction results. Oncological variables affected the RP decisions most, whereas RT was highly affected by geographic factors. The dependence plot depicted the feature interactions in reaching a treatment decision. The decision predicting model achieved an overall multiclass AUC of 0.77, whereas 0.74 was confirmed for the low-risk model. Using a large population-based real-world database, we unraveled the complex decision-making process and visualized nonlinear feature interactions in localized PCa.

A 'Portfolio of Model Approximations' approach to understanding invasion success with vector-borne disease

The central challenge of mathematical modeling of real-world systems is to strike an appropriate balance between insightful abstraction and detailed accuracy. Models in mathematical epidemiology frequently tend to either extreme, focusing on analytically provable boundaries in simplified, mass-action approximations, or else relying on calculated numerical solutions and computational simulation experiments to capture nuance and details specific to a particular host-disease system. We propose that there is value in an approach striking a slightly different compromise in which a detailed but analytically difficult system is modeled with careful detail, but then abstraction is applied to the results of numerical solutions to that system, rather than to the biological system itself. In this 'Portfolio of Model Approximations' approach, multiple levels of approximation are used to analyze the model at different scales of complexity. While this method has the potential to introduce error in the translation from model to model, it also has the potential to produce generalizable insight for the set of all similar systems, rather than isolated, tailored results that must be started anew for each next question. In this paper, we demonstrate this process and its value with a case study from evolutionary epidemiology. We consider a modified Susceptible-Infected-Recovered model for a vector-borne pathogen affecting two annually reproducing hosts. From observing patterns in simulations of the system and exploiting basic epidemiological properties, we construct two approximations of the model at different levels of complexity that can be treated as hypotheses about the behavior of the model. We compare the predictions of the approximations to the simulated results and discuss the trade-offs between accuracy and abstraction. We discuss the implications for this particular model, and in the context of mathematical biology in general.

The role of machine learning in HIV risk prediction

Despite advances in reducing HIV-related mortality, persistently high HIV incidence rates are undermining global efforts to end the epidemic by 2030. The UNAIDS Fast-track targets as well as other preventative strategies, such as pre-exposure prophylaxis, have been identified as priority areas to reduce the ongoing transmission threatening to undermine recent progress. Accurate and granular risk prediction is critical for these campaigns but is often lacking in regions where the burden is highest. Owing to their ability to capture complex interactions between data, machine learning and artificial intelligence algorithms have proven effective at predicting the risk of HIV infection in both high resource and low resource settings. However, interpretability of these algorithms presents a challenge to the understanding and adoption of these algorithms. In this perspectives article, we provide an introduction to machine learning and discuss some of the important considerations when choosing the variables used in model development and when evaluating the performance of different machine learning algorithms, as well as the role emerging tools such as Shapely Additive Explanations may play in helping understand and decompose these models in the context of HIV. Finally, we discuss some of the potential public health and clinical use cases for such decomposed risk assessment models in directing testing and preventative interventions including pre-exposure prophylaxis, as well as highlight the potential integration synergies with algorithms that predict the risk of sexually transmitted infections and tuberculosis.

Through the looking glass: Deep interpretable dynamic directed connectivity in resting fMRI

Brain network interactions are commonly assessed via functional (network) connectivity, captured as an undirected matrix of Pearson correlation coefficients. Functional connectivity can represent static and dynamic relations, but often these are modeled using a fixed choice for the data window Alternatively, deep learning models may flexibly learn various representations from the same data based on the model architecture and the training task. However, the representations produced by deep learning models are often difficult to interpret and require additional posthoc methods, e.g., saliency maps. In this work, we integrate the strengths of deep learning and functional connectivity methods while also mitigating their weaknesses. With interpretability in mind, we present a deep learning architecture that exposes a directed graph layer that represents what the model has learned about relevant brain connectivity. A surprising benefit of this architectural interpretability is significantly improved accuracy in discriminating controls and patients with schizophrenia, autism, and dementia, as well as age and gender prediction from functional MRI data. We also resolve the window size selection problem for dynamic directed connectivity estimation as we estimate windowing functions from the data, capturing what is needed to estimate the graph at each time-point. We demonstrate efficacy of our method in comparison with multiple existing models that focus on classification accuracy, unlike our interpretability-focused architecture. Using the same data but training different models on their own discriminative tasks we are able to estimate task-specific directed connectivity matrices for each subject. Results show that the proposed approach is also more robust to confounding factors compared to standard dynamic functional connectivity models. The dynamic patterns captured by our model are naturally interpretable since they highlight the intervals in the signal that are most important for the prediction. The proposed approach reveals that differences in connectivity among sensorimotor networks relative to default-mode networks are an important indicator of dementia and gender. Dysconnectivity between networks, specially sensorimotor and visual, is linked with schizophrenic patients, however schizophrenic patients show increased intra-network default-mode connectivity compared to healthy controls. Sensorimotor connectivity was important for both dementia and schizophrenia prediction, but schizophrenia is more related to dysconnectivity between networks whereas, dementia bio-markers were mostly intra-network connectivity.

Camel (Camelus spp.) Urine Bioactivity and Metabolome: A Systematic Review of Knowledge Gaps, Advances, and Directions for Future Research

Up to the present day, studies on the therapeutic properties of camel (Camelus spp.) urine and the detailed characterization of its metabolomic profile are scarce and often unrelated. Information on inter individual variability is noticeably limited, and there is a wide divergence across studies regarding the methods for sample storage, pre-processing, and extract derivatization for metabolomic analysis. Additionally, medium osmolarity is not experimentally adjusted prior to bioactivity assays. In this scenario, the methodological standardization and interdisciplinary approach of such processes will strengthen the interpretation, repeatability, and replicability of the empirical results on the compounds with bioactive properties present in camel urine. Furthermore, sample enlargement would also permit the evaluation of camel urine's intra- and interindividual variability in terms of chemical composition, bioactive effects, and efficacy, while it may also permit researchers to discriminate potential animal-intrinsic and extrinsic conditioning factors. Altogether, the results would help to evaluate the role of camel urine as a natural source for the identification and extraction of specific novel bioactive substances that may deserve isolated chemical and pharmacognostic investigations through preclinical tests to determine their biological activity and the suitability of their safety profile for their potential inclusion in therapeutic formulas for improving human and animal health.

Machine learning did not beat logistic regression in time series prediction for severe asthma exacerbations

Early detection of severe asthma exacerbations through home monitoring data in patients with stable mild-to-moderate chronic asthma could help to timely adjust medication. We evaluated the potential of machine learning methods compared to a clinical rule and logistic regression to predict severe exacerbations. We used daily home monitoring data from two studies in asthma patients (development: n = 165 and validation: n = 101 patients). Two ML models (XGBoost, one class SVM) and a logistic regression model provided predictions based on peak expiratory flow and asthma symptoms. These models were compared with an asthma action plan rule. Severe exacerbations occurred in 0.2% of all daily measurements in the development (154/92,787 days) and validation cohorts (94/40,185 days). The AUC of the best performing XGBoost was 0.85 (0.82-0.87) and 0.88 (0.86-0.90) for logistic regression in the validation cohort. The XGBoost model provided overly extreme risk estimates, whereas the logistic regression underestimated predicted risks. Sensitivity and specificity were better overall for XGBoost and logistic regression compared to one class SVM and the clinical rule. We conclude that ML models did not beat logistic regression in predicting short-term severe asthma exacerbations based on home monitoring data. Clinical application remains challenging in settings with low event incidence and high false alarm rates with high sensitivity.

Feasibility Study of Constructing a Screening Tool for Adolescent Diabetes Detection Applying Machine Learning Methods

Prediabetes and diabetes are becoming alarmingly prevalent among adolescents over the past decade. However, an effective screening tool that can assess diabetes risks smoothly is still in its infancy. In order to contribute to such significant gaps, this research proposes a machine learning-based predictive model to detect adolescent diabetes. The model applies supervised machine learning and a novel feature selection method to the National Health and Nutritional Examination Survey datasets after an exhaustive search to select reliable and accurate data. The best model achieved an area under the curve (AUC) score of 71%. This research proves that a screening tool based on supervised machine learning models can assist in the automated detection of youth diabetes. It also identifies some critical predictors to such detection using Lasso Regression, Random Forest Importance and Gradient Boosted Tree Importance feature selection methods. The most contributing features to Youth diabetes detection are physical characteristics (e.g., waist, leg length, gender), dietary information (e.g., water, protein, sodium) and demographics. These predictors can be further utilised in other areas of medical research, such as electronic medical history.

Review of applications and challenges of quantitative systems pharmacology modeling and machine learning for heart failure

Quantitative systems pharmacology (QSP) is an important approach in pharmaceutical research and development that facilitates in silico generation of quantitative mechanistic hypotheses and enables in silico trials. As demonstrated by applications from numerous industry groups and interest from regulatory authorities, QSP is becoming an increasingly critical component in clinical drug development. With rapidly evolving computational tools and methods, QSP modeling has achieved important progress in pharmaceutical research and development, including for heart failure (HF). However, various challenges exist in the QSP modeling and clinical characterization of HF. Machine/deep learning (ML/DL) methods have had success in a wide variety of fields and disciplines. They provide data-driven approaches in HF diagnosis and modeling, and offer a novel strategy to inform QSP model development and calibration. The combination of ML/DL and QSP modeling becomes an emergent direction in the understanding of HF and clinical development new therapies. In this work, we review the current status and achievement in QSP and ML/DL for HF, and discuss remaining challenges and future perspectives in the field.

Effect of AI Explanations on Human Perceptions of Patient-Facing AI-Powered Healthcare Systems

Ongoing research efforts have been examining how to utilize artificial intelligence technology to help healthcare consumers make sense of their clinical data, such as diagnostic radiology reports. How to promote the acceptance of such novel technology is a heated research topic. Recent studies highlight the importance of providing local explanations about AI prediction and model performance to help users determine whether to trust AI's predictions. Despite some efforts, limited empirical research has been conducted to quantitatively measure how AI explanations impact healthcare consumers' perceptions of using patient-facing, AI-powered healthcare systems. The aim of this study is to evaluate the effects of different AI explanations on people's perceptions of AI-powered healthcare system. In this work, we designed and deployed a large-scale experiment (N = 3,423) on Amazon Mechanical Turk (MTurk) to evaluate the effects of AI explanations on people's perceptions in the context of comprehending radiology reports. We created four groups based on two factors-the extent of explanations for the prediction (High vs. Low Transparency) and the model performance (Good vs. Weak AI Model)-and randomly assigned participants to one of the four conditions. Participants were instructed to classify a radiology report as describing a normal or abnormal finding, followed by completing a post-study survey to indicate their perceptions of the AI tool. We found that revealing model performance information can promote people's trust and perceived usefulness of system outputs, while providing local explanations for the rationale of a prediction can promote understandability but not necessarily trust. We also found that when model performance is low, the more information the AI system discloses, the less people would trust the system. Lastly, whether human agrees with AI predictions or not and whether the AI prediction is correct or not could also influence the effect of AI explanations. We conclude this paper by discussing implications for designing AI systems for healthcare consumers to interpret diagnostic report.

Could graph neural networks learn better molecular representation for drug discovery? A comparison study of descriptor-based and graph-based models

Graph neural networks (GNN) has been considered as an attractive modelling method for molecular property prediction, and numerous studies have shown that GNN could yield more promising results than traditional descriptor-based methods. In this study, based on 11 public datasets covering various property endpoints, the predictive capacity and computational efficiency of the prediction models developed by eight machine learning (ML) algorithms, including four descriptor-based models (SVM, XGBoost, RF and DNN) and four graph-based models (GCN, GAT, MPNN and Attentive FP), were extensively tested and compared. The results demonstrate that on average the descriptor-based models outperform the graph-based models in terms of prediction accuracy and computational efficiency. SVM generally achieves the best predictions for the regression tasks. Both RF and XGBoost can achieve reliable predictions for the classification tasks, and some of the graph-based models, such as Attentive FP and GCN, can yield outstanding performance for a fraction of larger or multi-task datasets. In terms of computational cost, XGBoost and RF are the two most efficient algorithms and only need a few seconds to train a model even for a large dataset. The model interpretations by the SHAP method can effectively explore the established domain knowledge for the descriptor-based models. Finally, we explored use of these models for virtual screening (VS) towards HIV and demonstrated that different ML algorithms offer diverse VS profiles. All in all, we believe that the off-the-shelf descriptor-based models still can be directly employed to accurately predict various chemical endpoints with excellent computability and interpretability.

Enabling personalized decision support with patient-generated data and attributable components

Decision-making related to health is complex. Machine learning (ML) and patient generated data can identify patterns and insights at the individual level, where human cognition falls short, but not all ML-generated information is of equal utility for making health-related decisions. We develop and apply attributable components analysis (ACA), a method inspired by optimal transport theory, to type 2 diabetes self-monitoring data to identify patterns of association between nutrition and blood glucose control. In comparison with linear regression, we found that ACA offers a number of characteristics that make it promising for use in decision support applications. For example, ACA was able to identify non-linear relationships, was more robust to outliers, and offered broader and more expressive uncertainty estimates. In addition, our results highlight a tradeoff between model accuracy and interpretability, and we discuss implications for ML-driven decision support systems.

Machine learning-based approaches for disease gene prediction

Disease gene prediction is an essential issue in biomedical research. In the early days, annotation-based approaches were proposed for this problem. With the development of high-throughput technologies, interaction data between genes/proteins have grown quickly and covered almost genome and proteome; thus, network-based methods for the problem become prominent. In parallel, machine learning techniques, which formulate the problem as a classification, have also been proposed. Here, we firstly show a roadmap of the machine learning-based methods for the disease gene prediction. In the beginning, the problem was usually approached using a binary classification, where positive and negative training sample sets are comprised of disease genes and non-disease genes, respectively. The disease genes are ones known to be associated with diseases; meanwhile, non-disease genes were randomly selected from those not yet known to be associated with diseases. However, the later may contain unknown disease genes. To overcome this uncertainty of defining the non-disease genes, more realistic approaches have been proposed for the problem, such as unary and semi-supervised classification. Recently, more advanced methods, including ensemble learning, matrix factorization and deep learning, have been proposed for the problem. Secondly, 12 representative machine learning-based methods for the disease gene prediction were examined and compared in terms of prediction performance and running time. Finally, their advantages, disadvantages, interpretability and trust were also analyzed and discussed.

How feasible is it to abandon statistical significance? A reflection based on a short survey

Background

There is a growing trend in using the “statistically significant” term in the scientific literature. However, harsh criticism of this concept motivated the recommendation to withdraw its use of scientific publications. We aimed to validate the support and the feasibility of adherence to this recommendation, among researchers having declared in favor of removing the statistical significance.

Methods

We surveyed signatories of an article published that defended this recommendation, to validate their opinion and ask them about how likely they will retire the concept of statistical significance.

Results

We obtained 151 responses which confirmed the support for the mentioned publication in aspects such as the adequate interpretation of the p-value, the degree of agreement, and the motivations to sign it. However, there was a wide distribution of answers about how likely are they to use the concept of “statistical significance” in future publications. About 42% declared being neutral, or that would likely use it again. We described arguments referred by several signatories and discussed aspects to be considered in the interpretation of research results.

Conclusions

The responses obtained from a proportion of signatories validated their declared position against the use of statistical significance. However, even in this group, the full application of this recommendation does not seem feasible. The arguments related to the inappropriate use of statistical tests should promote more education among researchers and users of scientific evidence.

Learning Molecular Representations for Medicinal Chemistry

The accurate modeling and prediction of small molecule properties and bioactivities depend on the critical choice of molecular representation. Decades of informatics-driven research have relied on expert-designed molecular descriptors to establish quantitative structure-activity and structure-property relationships for drug discovery. Now, advances in deep learning make it possible to efficiently and compactly learn molecular representations directly from data. In this review, we discuss how active research in molecular deep learning can address limitations of current descriptors and fingerprints while creating new opportunities in cheminformatics and virtual screening. We provide a concise overview of the role of representations in cheminformatics, key concepts in deep learning, and argue that learning representations provides a way forward to improve the predictive modeling of small molecule bioactivities and properties.

Interpretation of machine learning models using shapley values: application to compound potency and multi-target activity predictions

Difficulties in interpreting machine learning (ML) models and their predictions limit the practical applicability of and confidence in ML in pharmaceutical research. There is a need for agnostic approaches aiding in the interpretation of ML models regardless of their complexity that is also applicable to deep neural network (DNN) architectures and model ensembles. To these ends, the SHapley Additive exPlanations (SHAP) methodology has recently been introduced. The SHAP approach enables the identification and prioritization of features that determine compound classification and activity prediction using any ML model. Herein, we further extend the evaluation of the SHAP methodology by investigating a variant for exact calculation of Shapley values for decision tree methods and systematically compare this variant in compound activity and potency value predictions with the model-independent SHAP method. Moreover, new applications of the SHAP analysis approach are presented including interpretation of DNN models for the generation of multi-target activity profiles and ensemble regression models for potency prediction.

Machine Learning in Psychometrics and Psychological Research

Recent controversies about the level of replicability of behavioral research analyzed using statistical inference have cast interest in developing more efficient techniques for analyzing the results of psychological experiments. Here we claim that complementing the analytical workflow of psychological experiments with Machine Learning-based analysis will both maximize accuracy and minimize replicability issues. As compared to statistical inference, ML analysis of experimental data is model agnostic and primarily focused on prediction rather than inference. We also highlight some potential pitfalls resulting from adoption of Machine Learning based experiment analysis. If not properly used it can lead to over-optimistic accuracy estimates similarly observed using statistical inference. Remedies to such pitfalls are also presented such and building model based on cross validation and the use of ensemble models. ML models are typically regarded as black boxes and we will discuss strategies aimed at rendering more transparent the predictions.

Interpretation of Compound Activity Predictions from Complex Machine Learning Models Using Local Approximations and Shapley Values

In qualitative or quantitative studies of structure-activity relationships (SARs), machine learning (ML) models are trained to recognize structural patterns that differentiate between active and inactive compounds. Understanding model decisions is challenging but of critical importance to guide compound design. Moreover, the interpretation of ML results provides an additional level of model validation based on expert knowledge. A number of complex ML approaches, especially deep learning (DL) architectures, have distinctive black-box character. Herein, a locally interpretable explanatory method termed Shapley additive explanations (SHAP) is introduced for rationalizing activity predictions of any ML algorithm, regardless of its complexity. Models resulting from random forest (RF), nonlinear support vector machine (SVM), and deep neural network (DNN) learning are interpreted, and structural patterns determining the predicted probability of activity are identified and mapped onto test compounds. The results indicate that SHAP has high potential for rationalizing predictions of complex ML models.

Chemical predictive modelling to improve compound quality

The 'quality' of small-molecule drug candidates, encompassing aspects including their potency, selectivity and ADMET (absorption, distribution, metabolism, excretion and toxicity) characteristics, is a key factor influencing the chances of success in clinical trials. Importantly, such characteristics are under the control of chemists during the identification and optimization of lead compounds. Here, we discuss the application of computational methods, particularly quantitative structure-activity relationships (QSARs), in guiding the selection of higher-quality drug candidates, as well as cultural factors that may have affected their use and impact.