Natural products (NPs) are primarily recognized as privileged structures to interact with protein drug targets.

Reducing hurdles to clinical trials without compromising the therapeutic promises of peptide candidates becomes an essential step in peptide-based drug design. Machine-learning models are cost-effective and time-saving strategies used to predict biological activities from primary sequences. Their limitations lie in the diversity of peptide sequences and biological information within these models. Additional outlier detection methods are needed to set the boundaries for reliable predictions; the applicability domain. Antimicrobial peptides (AMPs) constitute an extensive library of peptides offering promising avenues against antibiotic-resistant infections. Most AMPs present in clinical trials are administrated topically due to their hemolytic toxicity. Here we developed machine learning models and outlier detection methods that ensure robust predictions for the discovery of AMPs and the design of novel peptides with reduced hemolytic activity. Our best models, gradient boosting classifiers, predicted the hemolytic nature from any peptide sequence with 95–97% accuracy. Nearly 70% of AMPs were predicted as hemolytic peptides. Applying multivariate outlier detection models, we found that 273 AMPs (~ 9%) could not be predicted reliably. Our combined approach led to the discovery of 34 high-confidence non-hemolytic natural AMPs, the de novo design of 507 non-hemolytic peptides, and the guidelines for non-hemolytic peptide design.

Peptides have positively impacted the pharmaceutical industry as drugs, biomarkers, or diagnostic tools of high therapeutic value. However, only a handful have progressed to the market. Toxicity is one of the main obstacles to translating peptides into clinics. Hemolysis or hemotoxicity, the principal source of toxicity, is a natural or disease-induced event leading to the death of vital red blood cells. Initial screenings for toxicity have been widely evaluated using erythrocytes as the gold standard. More recently, many online databases filled with peptide sequences and their biological meta-data have paved the way toward hemolysis prediction using user-friendly, fast-access machine learning-driven programs. This review details the growing contributions of in silico approaches developed in the last decade for the large-scale prediction of erythrocyte lysis induced by peptides. After an overview of the pharmaceutical landscape of peptide therapeutics, we highlighted the relevance of early hemolysis studies in drug development. We emphasized the computational models and algorithms used to this end in light of historical and recent findings in this promising field. We benchmarked seven predictors using peptides from different data sets, having 7–35 amino acids in length. According to our predictions, the models have scored an accuracy over 50.42% and a minimal Matthew’s correlation coefficient over 0.11. The maximum values for these statistical parameters achieved 100.0% and 1.00, respectively. Finally, strategies for optimizing peptide selectivity were described, as well as prospects for future investigations. The development of in silico predictive approaches to peptide toxicity has just started, but their important contributions clearly demonstrate their potential for peptide science and computer-aided drug design. Methodology refinement and increasing use will motivate the timely and accurate in silico identification of selective, non-toxic peptide therapeutics.

The recent success of small-molecule kinase inhibitors as anticancer drugs has generated significant interest in their application to other clinical areas, such as disorders of the central nervous system (CNS). However, most kinase inhibitor drug candidates investigated to date have been ineffective at treating CNS disorders, mainly due to poor blood–brain barrier (BBB) permeability. It is, therefore, imperative to evaluate new chemical entities for both kinase inhibition and BBB permeability. Over the last 35 years, marine biodiscovery has yielded 471 natural products reported as kinase inhibitors, yet very few have been evaluated for BBB permeability. In this study, we revisited these marine natural products and predicted their ability to cross the BBB by applying freely available open-source chemoinformatics and machine learning algorithms to a training set of 332 previously reported CNS-penetrant small molecules. We evaluated several regression and classification models, and found that our optimised classifiers (random forest, gradient boosting, and logistic regression) outperformed other models, with overall cross-validated model accuracies of 80%–82% and 78%–80% on external testing. All 3 binary classifiers predicted 13 marine-derived kinase inhibitors with appropriate physicochemical characteristics for BBB permeability.

Machine learning models are revolutionizing our approaches to discovering and designing bioactive peptides. These models often need protein structure awareness, as they heavily rely on sequential data. The models excel at identifying sequences of a particular biological nature or activity, but they frequently fail to comprehend their intricate mechanism(s) of action. To solve two problems at once, we studied the mechanisms of action and structural landscape of antimicrobial peptides as (i) membrane-disrupting peptides, (ii) membrane-penetrating peptides, and (iii) protein-binding peptides. By analyzing critical features such as dipeptides and physicochemical descriptors, we developed models with high accuracy (86–88%) in predicting these categories. However, our initial models (1.0 and 2.0) exhibited a bias towards α-helical and coiled structures, influencing predictions. To address this structural bias, we implemented subset selection and data reduction strategies. The former gave three structure-specific models for peptides likely to fold into α-helices (models 1.1 and 2.1), coils (1.3 and 2.3), or mixed structures (1.4 and 2.4). The latter depleted over-represented structures, leading to structure-agnostic predictors 1.5 and 2.5. Additionally, our research highlights the sensitivity of important features to different structure classes across models.