Machine Learning Methods in Drug Discovery

The advancements of information technology and related processing techniques have created a fertile base for progress in many scientific fields and industries. In the fields of drug discovery and development, machine learning techniques have been used for the development of novel drug candidates. The methods for designing drug targets and novel drug discovery now routinely combine machine learning and deep learning algorithms to enhance the efficiency, efficacy, and quality of developed outputs. The generation and incorporation of big data, through technologies such as high-throughput screening and high through-put computational analysis of databases used for both lead and target discovery, has increased the reliability of the machine learning and deep learning incorporated techniques. The use of these virtual screening and encompassing online information has also been highlighted in developing lead synthesis pathways. In this review, machine learning and deep learning algorithms utilized in drug discovery and associated techniques will be discussed. The applications that produce promising results and methods will be reviewed.     

Alkaloids Used as Medicines: Structural Phytochemistry Meets Biodiversity—An Update and Forward Look

Selecting candidates for drug developments using computational design and empirical rules has resulted in a broad discussion about their success. In a previous study, we had shown that a species’ abundance [as expressed by the GBIF (Global Biodiversity Information Facility)] dataset is a core determinant for the development of a natural product into a medicine. Our overarching aim is to understand the unique requirements for natural product-based drug development. Web of Science was queried for research on alkaloids in combination with plant systematics/taxonomy. All alkaloids containing species demonstrated an average increase of 8.66 in GBIF occurrences between 2014 and 2020. Medicinal Species with alkaloids show higher abundance compared to non-medicinal alkaloids, often linked also to cultivation. Alkaloids with high biodiversity are often simple alkaloids found in multiple species with the presence of ’driver species‘ and are more likely to be included in early-stage drug development compared to ‘rare’ alkaloids. Similarly, the success of an alkaloid containing species as a food supplement (‘botanical’) is linked to its abundance. GBIF is a useful tool for assessing the druggability of a compound from a certain source species. The success of any development programme from natural sources must take sustainable sourcing into account right from the start.     

Approach for the Design of Covalent Protein Kinase Inhibitors via Focused Deep Generative Modeling

Deep machine learning is expanding the conceptual framework and capacity of computational compound design, enabling new applications through generative modeling. We have explored the systematic design of covalent protein kinase inhibitors by learning from kinome-relevant chemical space, followed by focusing on an exemplary kinase of interest. Covalent inhibitors experience a renaissance in drug discovery, especially for targeting protein kinases. However, computational design of this class of inhibitors has thus far only been little investigated. To this end, we have devised a computational approach combining fragment-based design and deep generative modeling augmented by three-dimensional pharmacophore screening. This approach is thought to be particularly relevant for medicinal chemistry applications because it combines knowledge-based elements with deep learning and is chemically intuitive. As an exemplary application, we report for Bruton’s tyrosine kinase (BTK), a major drug target for the treatment of inflammatory diseases and leukemia, the generation of novel candidate inhibitors with a specific chemically reactive group for covalent modification, requiring only little target-specific compound information to guide the design efforts. Newly generated compounds include known inhibitors and characteristic substructures and many novel candidates, thus lending credence to the computational approach, which is readily applicable to other targets.     

Automated discovery of noncovalent inhibitors of SARS-CoV-2 main protease by consensus Deep Docking of 40 billion small molecules

Recent explosive growth of ‘make-on-demand’ chemical libraries brought unprecedented opportunities but also significant challenges to the field of computer-aided drug discovery. To address this expansion of the accessible chemical universe, molecular docking needs to accurately rank billions of chemical structures, calling for the development of automated hit-selecting protocols to minimize human intervention and error. Herein, we report the development of an artificial intelligence-driven virtual screening pipeline that utilizes Deep Docking with Autodock GPU, Glide SP, FRED, ICM and QuickVina2 programs to screen 40 billion molecules against SARS-CoV-2 main protease (Mpro). This campaign returned a significant number of experimentally confirmed inhibitors of Mpro enzyme, and also enabled to benchmark the performance of twenty-eight hit-selecting strategies of various degrees of stringency and automation. These findings provide new starting scaffolds for hit-to-lead optimization campaigns against Mpro and encourage the development of fully automated end-to-end drug discovery protocols integrating machine learning and human expertise.

Deep learning-accelerated docking coupled with computational hit selection strategies enable the identification of inhibitors for the SARS-CoV-2 main protease from a chemical library of 40 billion small molecules.     

Virtual Screening with Gnina 1.0

Virtual screening—predicting which compounds within a specified compound library bind to a target molecule, typically a protein—is a fundamental task in the field of drug discovery. Doing virtual screening well provides tangible practical benefits, including reduced drug development costs, faster time to therapeutic viability, and fewer unforeseen side effects. As with most applied computational tasks, the algorithms currently used to perform virtual screening feature inherent tradeoffs between speed and accuracy. Furthermore, even theoretically rigorous, computationally intensive methods may fail to account for important effects relevant to whether a given compound will ultimately be usable as a drug. Here we investigate the virtual screening performance of the recently released Gnina molecular docking software, which uses deep convolutional networks to score protein-ligand structures. We find, on average, that Gnina outperforms conventional empirical scoring. The default scoring in Gnina outperforms the empirical AutoDock Vina scoring function on 89 of the 117 targets of the DUD-E and LIT-PCBA virtual screening benchmarks with a median 1% early enrichment factor that is more than twice that of Vina. However, we also find that issues of bias linger in these sets, even when not used directly to train models, and this bias obfuscates to what extent machine learning models are achieving their performance through a sophisticated interpretation of molecular interactions versus fitting to non-informative simplistic property distributions.     

Quantum algorithm for alchemical optimization in material design

The development of tailored materials for specific applications is an active field of research in chemistry, material science and drug discovery. The number of possible molecules obtainable from a set of atomic species grow exponentially with the size of the system, limiting the efficiency of classical sampling algorithms. On the other hand, quantum computers can provide an efficient solution to the sampling of the chemical compound space for the optimization of a given molecular property. In this work, we propose a quantum algorithm for addressing the material design problem with a favourable scaling. The core of this approach is the representation of the space of candidate structures as a linear superposition of all possible atomic compositions. The corresponding ‘alchemical’ Hamiltonian drives the optimization in both the atomic and electronic spaces leading to the selection of the best fitting molecule, which optimizes a given property of the system, e.g., the interaction with an external potential as in drug design. The quantum advantage resides in the efficient calculation of the electronic structure properties together with the sampling of the exponentially large chemical compound space. We demonstrate both in simulations and with IBM Quantum hardware the efficiency of our scheme and highlight the results in a few test cases. This preliminary study can serve as a basis for the development of further material design quantum algorithms for near-term quantum computers.

‘Alchemical’ quantum algorithm for the simultaneous optimisation of chemical composition and electronic structure for material design. By exploiting quantum mechanical principles this approach will boost drug discovery in the near future.     

High Throughput Identification of the Potential Antioxidant Peptides in Ophiocordyceps sinensis

Ophiocordyceps sinensis, an ascomycete caterpillar fungus, has been used as a Traditional Chinese Medicine owing to its bioactive properties. However, until now the bio-active peptides have not been identified in this fungus. Here, the raw RNA sequences of three crucial growth stages of the artificially cultivated O. sinensis and the wild-grown mature fruit-body were aligned to the genome of O. sinensis. Both homology-based prediction and de novo-based prediction methods were used to identify 8541 putative antioxidant peptides (pAOPs). The expression profiles of the cultivated mature fruiting body were similar to those found in the wild specimens. The differential expression of 1008 pAOPs matched genes had the highest difference between ST and MF, suggesting that the pAOPs were primarily induced and play important roles in the process of the fruit-body maturation. Gene ontology analysis showed that most of pAOPs matched genes were enriched in terms of ‘cell redox homeostasis’, ‘response to oxidative stresses’, ‘catalase activity’, and ‘ integral component of cell membrane’. A total of 1655 pAOPs was identified in our protein-seqs, and some crucial pAOPs were selected, including catalase, peroxiredoxin, and SOD [Cu–Zn]. Our findings offer the first identification of the active peptide ingredients in O. sinensis, facilitating the discovery of anti-infectious bio-activity and the understanding of the roles of AOPs in fungal pathogenicity and the high-altitude adaptation in this medicinal fungus.     

Chemical Diversity and Potential Target Network of Woody Peony Flower Essential Oil from Eleven Representative Cultivars (Paeonia × suffruticosa Andr.)

Woody peony (Paeonia × suffruticosa Andr.) has many cultivars with genetic variances. The flower essential oil is valued in cosmetics and fragrances. This study was to investigate the chemical diversity of essential oils of eleven representative cultivars and their potential target network. Hydro-distillation afforded yields of 0.11–0.25%. Essential oils were analyzed by GC-MS and GC-FID which identified 105 compounds. Three clusters emerged from multivariate analysis, representative of phloroglucinol trimethyl ether (‘Caihui’), citronellol (‘Jingyu’, ‘Zhaofen’ and ‘Baiyuan Zhenghui’) and mixed (the rest of the cultivars) chemotypes. ‘Zhaofen’ and ‘Jingyu’ also exhibited low levels of other rose-related compounds. The main components were subjected to a target network approach. Drug-likeness screening gave 20 compounds with predictive blood–brain barrier permeation. Compound target network identified six key compounds, namely nerol, citronellol, geraniol, geranic acid, cis-3-hexen-1-ol and 1-hexanol. Top enriched terms in GO, KEGG and DisGeNET were mostly related to the central nervous system (CNS). Protein—protein interactions revealed a core network of 14 targets, 11 of which were CNS-related (targets for antidepressants, analgesics, antipsychotics, anti-Alzheimer’s and anti-Parkinson’s agents). This work provides useful information on the production of woody peony essential oils with specific chemotypes and reveals their potential importance in aromatherapy for alternative treatment of CNS disorders.     

A machine learning approach to computer-aided molecular design

Summary Preliminary results of a machine learning application concerning computer-aided molecular design applied to drug discovery are presented. The artificial intelligence techniques of machine learning use a sample of active and inactive compounds, which is viewed as a set of positive and negative examples, to allow the induction of a molecular model characterizing the interaction between the compounds and a target molecule. The algorithm is based on a twofold phase. In the first one — the specialization step — the program identifies a number of active/inactive pairs of compounds which appear to be the most useful in order to make the learning process as effective as possible and generates a dictionary of molecular fragments, deemed to be responsible for the activity of the compounds. In the second phase — the generalization step — the fragments thus generated are combined and generalized in order to select the most plausible hypothesis with respect to the sample of compounds. A knowledge base concerning physical and chemical properties is utilized during the inductive process.     

Reuse of Food Waste: The Chemical Composition and Health Properties of Pomelo (Citrus maxima) Cultivar Essential Oils

The aim of the present study is to investigate the chemical profile, antioxidant activity, carbohydrate-hydrolysing enzyme inhibition, and hypolipidemic effect of essential oils (EOs) extracted from Sicilian Citrus maxima (pomelo) flavedo. Using gas-chromatography-mass spectrometry analysis (GC-MS) we analysed the Eos of five cultivars of C. maxima, namely, ‘Chadock’, ‘Maxima’, ‘Pyriformis’, ‘Terracciani’, and ‘Todarii’, and their blends. The antioxidant activity was performed by using a multi-target approach using 2,2′-Azino-Bis-3-Ethylbenzothiazoline-6-Sulfonic acid (ABTS), 2,2-Diphenyl-1-picrylhydrazyl (DPPH), ferric reducing ability power (FRAP), and β-carotene bleaching tests. The α-amylase, α-glucosidase, and lipase-inhibitory activities were also assessed. GC-MS analyses revealed D-limonene as the main monoterpene hydrocarbon in all cultivars, albeit with different percentages in the range of 21.72–71.13%. A good content of oxygenated monoterpenes was detected for all cultivars, especially for ‘Todarii’. The analysis of the principal components (PCA), and related clusters (HCA), was performed to find chemo-diversity among the analysed samples. EOs from ‘Chadock’ and ‘Maxima’ were statistically similar to each other, and they differed from P3 in the smaller amount of sesquiterpene hydrocarbons, while the oils from ‘Terracciani’ and ‘Todarii’ were found to be chemically and statistically different. ‘Chadock’ EO was the most active to scavenge radicals (IC₅₀ values of 22.24 and 27.23 µg/mL in ABTS and DPPH tests, respectively). ‘Terracciani’ EO was the most active against both lipase and α-amylase, whereas the blends obtained by the combination (1:1 v/v) of C. maxima ‘Maxima’ + ‘Todarii’ were the most active against α-glucosidase. Generally, the blends did not exert a unique behaviour in potentiating or reducing the bioactivity of the pomelo EOs.     

Sea Buckthorn Leaf Powders: The Impact of Cultivar and Drying Mode on Antioxidant,Phytochemical, and Chromatic Profile of Valuable Resource

Sea buckthorn (Hippophae rhamnoides L. (HR)) leaf powders are the underutilized, promising resource of valuable compounds. Genotype and processing methods are key factors in the preparation of homogenous, stable, and quantified ingredients. The aim of this study was to evaluate the phenolic, triterpenic, antioxidant profiles, carotenoid and chlorophyll content, and chromatic characteristics of convection-dried and freeze-dried HR leaf powders obtained from ten different female cultivars, namely ‘Avgustinka’, ‘Botaniceskaja Liubitelskaja’, ‘Botaniceskaja’, ‘Hibrid Percika’, ‘Julia’, ‘Nivelena’, ‘Otradnaja’, ‘Podarok Sadu’, ‘Trofimovskaja’, and ‘Vorobjovskaja’. The chromatic characteristics were determined using the CIELAB scale. The phytochemical profiles were determined using HPLC-PDA (high performance liquid chromatography with photodiode array detector) analysis; spectrophotometric assays and antioxidant activities were investigated using ABTS (2,2′-Azino-bis(3-ethylbenzothiazoline-6-sulfonic acid)) and FRAP (ferric ion reducing antioxidant power) assays. The sea buckthorn leaf powders had a yellowish-green appearance. The drying mode had a significant impact on the total antioxidant activity, chlorophyll content, and chromatic characteristics of the samples; the freeze-dried samples were superior in antioxidant activity, chlorophyll, carotenoid content, and chromatic profile, compared to convection-dried leaf powder samples. The determined triterpenic and phenolic profiles strongly depend on the cultivar, and the drying technique had no impact on qualitative and quantitative composition. Catechin, epigallocatechin, procyanidin B3, ursolic acid, α-amyrin, and β-sitosterol could be used as quantitative markers in the phenolic and triterpenic profiles. The cultivars ‘Avgustinka’, ‘Nivelena’, and ‘Botaniceskaja’ were superior to other tested cultivars, with the phytochemical composition and antioxidant activity.     

Plant Secondary Metabolites Produced in Response to Abiotic Stresses Has Potential Application in Pharmaceutical Product Development

Plant secondary metabolites (PSMs) are vital for human health and constitute the skeletal framework of many pharmaceutical drugs. Indeed, more than 25% of the existing drugs belong to PSMs. One of the continuing challenges for drug discovery and pharmaceutical industries is gaining access to natural products, including medicinal plants. This bottleneck is heightened for endangered species prohibited for large sample collection, even if they show biological hits. While cultivating the pharmaceutically interesting plant species may be a solution, it is not always possible to grow the organism outside its natural habitat. Plants affected by abiotic stress present a potential alternative source for drug discovery. In order to overcome abiotic environmental stressors, plants may mount a defense response by producing a diversity of PSMs to avoid cells and tissue damage. Plants either synthesize new chemicals or increase the concentration (in most instances) of existing chemicals, including the prominent bioactive lead compounds morphine, camptothecin, catharanthine, epicatechin-3-gallate (EGCG), quercetin, resveratrol, and kaempferol. Most PSMs produced under various abiotic stress conditions are plant defense chemicals and are functionally anti-inflammatory and antioxidative. The major PSM groups are terpenoids, followed by alkaloids and phenolic compounds. We have searched the literature on plants affected by abiotic stress (primarily studied in the simulated growth conditions) and their PSMs (including pharmacological activities) from PubMed, Scopus, MEDLINE Ovid, Google Scholar, Databases, and journal websites. We used search keywords: “stress-affected plants,” “plant secondary metabolites, “abiotic stress,” “climatic influence,” “pharmacological activities,” “bioactive compounds,” “drug discovery,” and “medicinal plants” and retrieved published literature between 1973 to 2021. This review provides an overview of variation in bioactive phytochemical production in plants under various abiotic stress and their potential in the biodiscovery of therapeutic drugs. We excluded studies on the effects of biotic stress on PSMs.     

Content of Phenolic Compounds and Organic Acids in the Flowers of Selected Tulipa gesneriana Cultivars

The study focused on the determination of phenolic acids, flavonoids and organic acids in five tulip cultivars ‘Barcelona’, ‘Columbus’, ‘Strong Gold’, ‘Super Parrot’ and ‘Tropicana’. The cultivars grown in field and in a greenhouse were exposed after cutting to different times of storage (0, 3 and 6 days). The phenolic profile contained 4-hydroxybenzoic, 2,5-dihydroxybenzoic, gallic, vanillic, syringic, salicylic, protocatechuic, trans-cinnamic, p-coumaric, caffeic, ferulic, chlorogenic and sinapic acids, as well as quercetin, rutin, luteonin, catechin and vitexin. The mean phenolic acid content was in the following order: ‘Columbus’ and ‘Tropicana’ > ’Barcelona’ > ’Strong Gold’ > ’Super Parrot’, while the levels of flavonoids were as follows: ‘Strong Gold’ > ’Barcelona’ > ’Tropicana’ > ’Columbus’ > ’Super Parrot’. The highest content of phenolic acids was confirmed for Columbus and Tropicana, while the lowest was for Super Parrot. However total phenolic content was very similar, observed between the place of cultivation, time of storage and cultivars. Malonic, succinic, acetic and citric acids were the major organic acid components in tulip petals. More organic acids (except malonic) were accumulated in tulip petals from fields than those from the greenhouse, while changes during storage were strictly correlated with cultivars.     

Drug Design—Past,Present, Future

Drug design is a complex pharmaceutical science with a long history. Many achievements have been made in the field of drug design since the end of 19th century, when Emil Fisher suggested that the drug–receptor interaction resembles the key and lock interplay. Gradually, drug design has been transformed into a coherent and well-organized science with a solid theoretical background and practical applications. Now, drug design is the most advanced approach for drug discovery. It utilizes the innovations in science and technology and includes them in its wide-ranging arsenal of methods and tools in order to achieve the main goal: discovery of effective, specific, non-toxic, safe and well-tolerated drugs. Drug design is one of the most intensively developing modern sciences and its progress is accelerated by the implication of artificial intelligence. The present review aims to capture some of the most important milestones in the development of drug design, to outline some of the most used current methods and to sketch the future perspective according to the author’s point of view. Without pretending to cover fully the wide range of drug design topics, the review introduces the reader to the content of Molecules’ Special Issue “Drug Design—Science and Practice”.     

Comparative Assessment of Phytochemical Compounds and Antioxidant Properties of Kernel Oil from Eight Sour Cherry (Prunus cerasus L.) Cultivars

New plant oils as a potential natural source of nutraceutical compounds are still being sought. The main components of eight cultivars (‘Koral’, ‘Lucyna’, ‘Montmorency’, ‘Naumburger’, ‘Wanda’, ‘Wigor’, ‘Wołyńska’, and ‘Wróble’) of sour cherry (Prunus cerasus L.) grown in Poland, including crude fat, protein, and oil content, were evaluated. The extracted oils were analysed for chemical and biological activity. The oils had an average peroxide value of 1.49 mEq O₂/kg, acid value of 1.20 mg KOH/g, a saponification value of 184 mg of KOH/g, and iodine value of 120 g I₂/100 g of oil. The sour cherry oil contained linoleic (39.1–46.2%) and oleic (25.4–41.0%) acids as the major components with smaller concentrations of α-eleostearic acid (8.00–15.62%), palmitic acid (5.45–7.41%), and stearic acid (2.49–3.17%). The content of sterols and squalene varied significantly in all the studied cultivars and ranged between 336–973 mg/100 g and 66–102 mg/100 g of oil. The contents of total tocochromanols, polyphenols, and carotenoids were 119–164, 19.6–29.5, and 0.56–1.61 mg/100 g oil, respectively. The cultivar providing the highest amounts of oil and characterised by the highest content of PUFA (including linoleic acid), plant sterols, α-and β-tocopherol, as well as the highest total polyphenol and total carotenoids content was been found to be ‘Naumburger’. The antioxidant capacity of sour cherry kernel oils, measured using the DPPH^• and ABTS^•+ methods, ranged from 57.7 to 63.5 and from 38.2 to 43.2 mg trolox/100 g oil, respectively. The results of the present study provide important information about potential possibilities of application of Prunus cerasus kernel oils in cosmetic products and pharmaceuticals offering health benefits.     

Unravelling the Anticancer Mechanisms of Traditional Herbal Medicines with Metabolomics

Metabolite profiling of cancer cells presents many opportunities for anticancer drug discovery. The Chinese, Indian, and African flora, in particular, offers a diverse source of anticancer therapeutics as documented in traditional folklores. In-depth scientific information relating to mechanisms of action, quality control, and safety profile will promote their extensive usage in cancer therapy. Metabolomics may be a more holistic strategy to gain valuable insights into the anticancer mechanisms of action of plants but this has remained largely unexplored. This review, therefore, presents the available metabolomics studies on the anticancer effects of herbal medicines commonly used in Africa and Asia. In addition, we present some scientifically understudied ‘candidate plants’ for cancer metabolomics studies and highlight the relevance of metabolomics in addressing other challenges facing the drug development of anticancer herbs. Finally, we discussed the challenges of using metabolomics to uncover the underlying mechanisms of potential anticancer herbs and the progress made in this regard.     

A Comparison between Different Agro-wastes and Carbon Nanotubes for Removal of Sarafloxacin from Wastewater: Kinetics and Equilibrium Studies

In the current study, eco-structured and efficient removal of the veterinary fluoroquinolone antibiotic sarafloxacin (SARA) from wastewater has been explored. The adsorptive power of four agro-wastes (AWs) derived from pistachio nutshells (PNS) and Aloe vera leaves (AV) as well as the multi-walled carbon nanotubes (MWCNTs) has been assessed. Adsorbent derived from raw pistachio nutshells (RPNS) was the most efficient among the four tested AWs (%removal ‘%R’ = 82.39%), while MWCNTs showed the best adsorptive power amongst the five adsorbents (%R = 96.20%). Plackett-Burman design (PBD) was used to optimize the adsorption process. Two responses (‘%R’ and adsorption capacity ‘q_e’) were optimized as a function of four variables (pH, adsorbent dose ‘AD’ (dose of RPNS and MWCNTs), adsorbate concentration [SARA] and contact time ‘CT’). The effect of pH was similar for both RPNS and MWCNTs. Morphological and textural characterization of the tested adsorbents was carried out using FT-IR spectroscopy, SEM and BET analyses. Conversion of waste-derived materials into carbonaceous material was investigated by Raman spectroscopy. Equilibrium studies showed that Freundlich isotherm is the most suitable isotherm to describe the adsorption of SARA onto RPNS. Kinetics’ investigation shows that the adsorption of SARA onto RPNS follows a pseudo-second order (PSO) model.     

Structural rearrangement of Neisseria meningitidis transferrin binding protein A (TbpA) prior to human transferrin protein (hTf) binding

Gram-negative bacterium Neisseria meningitidis, responsible for human infectious disease meningitis, acquires the iron (Fe³⁺) ion needed for its survival from human transferrin protein (hTf). For this transport, transferrin binding proteins TbpA and TbpB are facilitated by the bacterium. The transfer cannot occur without TbpA, while the absence of TbpB only slows down the transfer. Thus, understanding the TbpA-hTf binding at the atomic level is crucial for the fight against bacterial meningitis infections. In this study, atomistic level of mechanism for TbpA-hTf binding is elucidated through 100 ns long all-atom classical MD simulations on free (uncomplexed) TbpA. TbpA protein underwent conformational change from ‘open’ state to ‘closed’ state, where two loop domains, loops 5 and 8, were very close to each other. This state clearly cannot accommodate hTf in the cleft between these two loops. Moreover, the helix finger domain, which might play a critical role in Fe³⁺ ion uptake, also shifted downwards leading to unfavorable Tbp-hTf binding. Results of this study indicated that TbpA must switch between ‘closed’ state to ‘open’ state, where loops 5 and 8 are far from each other creating a cleft for hTf binding. The atomistic level of understanding to conformational switch is crucial for TbpA-hTf complex inhibition strategies. Drug candidates can be designed to prevent this conformational switch, keeping TbpA locked in ‘closed’ state.