Computational Molecular Biology 2024, Vol.14, No.3, 115-124 http://bioscipublisher.com/index.php/cmb 118 drug design, and the prediction of physicochemical and pharmacokinetic properties (Yang et al., 2019). These applications have expedited the early drug discovery process by predicting protein structures, drug-target interactions, and molecular properties such as drug toxicity. AI has also been instrumental in drug repurposing and optimizing drug design by leveraging large datasets and complex models(Figure 1) (Jiménez-Luna et al., 2021). Figure 1 Schematic diagram of the transition between classical and modern methodologies for some relevant problems in drug discovery, such as QSAR/QSPR modeling, de novo drug design, and synthesis planning. Abbreviations: ML, machine learning; SVM, support vector machine; RF, random forest; QSAR/QSPR, quantitative structure-activity/property relationship; NN, neural network; SE(3), special Euclidean group in three-dimensions; NLP, natural language processing; MCTS, Monte Carlo tree search (Adopted from Jiménez-Luna et al., 2021) The study of Jiménez-Luna et al. (2021) highlights the evolution from classical to modern approaches in drug discovery, particularly in QSAR/QSPR modeling, de novo drug design, and synthesis planning. Classical methods rely on handcrafted descriptors, linear models, and traditional machine learning techniques like SVMs and RFs. In contrast, modern methods integrate advanced neural networks, such as graph NNs, and natural language processing (NLP) techniques, enabling rule-free predictions, multitask learning, and latent-space optimization. This transition represents the growing influence of artificial intelligence (AI) in enhancing the accuracy, efficiency, and scalability of drug design and synthesis processes.
RkJQdWJsaXNoZXIy MjQ4ODYzNA==