This Review explores the state-of-the-art applications of AI in small-molecule drug development since 2019. It first describes AI-powered drug discovery, encompassing target identification, synthesis planning, and clinical stages of drug development. Applications include biomarker discovery, drug repurposing, and prediction of pharmacokinetic properties.
AI has transformed drug discovery methodologies, enhancing efficiency across various stages. It facilitates target identification, virtual screening, ADMET (absorption, distribution, metabolism, excretion, and toxicity) predictions, and synthesis automation. Researchers leverage advanced algorithms to accelerate the discovery of novel therapeutic agents, improving prediction accuracy and reducing costs.
Identifying small-molecule targets is critical in drug discovery. Traditional methods, such as affinity pull-down and whole-genome knockdown screening, are time-consuming and labor-intensive. AI enables analysis of large datasets within complex biological networks, identifying disease-related molecular patterns and candidate drug targets.
Recent research employs NLP techniques, like word2vec embeddings, to enhance target identification sensitivity. Graph deep learning technology merges graph structures with deep learning, effectively identifying candidate targets. A recent study developed an interpretable framework using multi-omics network graphs to predict cancer genes.
Integrating multi-omics data with scientific literature into knowledge graphs allows AI to discern relationships between genes and disease pathways. Biomedical LLMs, integrated with biological networks, provide efficient methods for linking diseases, genes, and biological processes. The PandaOmics platform recognized TRAF2- and NCK-interacting kinase as a potential target for anti-fibrotic therapy, leading to the development of a specific inhibitor.
Real-world data, such as medical records and electronic health records, provide essential context for understanding complex diseases. However, unstructured text and lack of standardization limit their application. Despite challenges, recent studies show that noisy real-world data can train effective models for gene discovery.
Virtual screening is crucial for identifying potential drug candidates. AI-based receptor-ligand docking models predict ligand transformations and generate atomic coordinates. Recent advancements in receptor-ligand co-folding networks show promise in predicting complex structures directly from sequence information.
In the absence of target structures, AI techniques can be used in sequence-based prediction methods, although these often struggle with the complexity of protein-ligand interactions. Phenotype-based virtual screening is vital for diseases lacking defined targets.
Current models focus on specific tasks, emphasizing the need for universal models capable of handling multiple tasks. Active learning and Bayesian optimization are effective methods for enhancing virtual screening efficiency. Integrating quantum mechanics with AI offers new tools for chemical space exploration.
De novo drug design, enabled by AI, autonomously creates new chemical structures meeting desired molecular features. Traditional methods rely on expert designers. AI has automated the identification of novel structures, leading to a more efficient drug discovery era.
ADMET predictions are critical for determining drug efficacy and safety. AI predicts ADMET properties using predefined features, with deep learning now driving these predictions. Despite advances, challenges remain, including high costs and limited labeled data.
Interpretability in AI models is essential for understanding relationships between molecular substructures and properties. Attention mechanisms enhance interpretability by identifying key atoms. The integration of chemical knowledge can further improve model transparency.
Automated synthesis of organic compounds represents a cutting-edge frontier in chemistry. CASP assists chemists in determining reaction routes. Recent advancements in deep learning-powered automated synthesis techniques have improved efficiency in drug discovery.
AI technology facilitates the in vivo validation of new drug mechanisms of action. High-content screening, monitoring real-time changes in omics data, allows AI to develop models capable of deciphering the mechanisms and properties of new compounds.