Abstract :
The integration of artificial intelligence (AI) with quantitative structure–activity relationship (QSAR) modeling has transformed drug discovery and medicinal chemistry by enabling rapid, accurate, and large-scale prediction of molecular properties and bioactivities. Traditional QSAR approaches are limited by linear assumptions, small datasets, and limited feature extraction capabilities. In contrast, modern deep-learning (DL) frameworks such as convolutional neural networks, graph neural networks, and recurrent architectures can automatically extract high-level chemical descriptors, handle complex nonlinearities, and generalize across diverse chemical spaces. This article critically reviews the evolution of QSAR from classical statistical models to AI-driven deep-learning paradigms, highlighting advances in molecular representation (e.g., SMILES embeddings, molecular graphs), model interpretability, and integration with cheminformatics workflows. We discuss recent applications of deep-learning QSAR in hit identification, lead optimization, ADMET prediction, and polypharmacology. Challenges such as data scarcity, overfitting, and the need for transparent, regulatory-compliant models are addressed. Finally, future directions are proposed for developing explainable and transferable AI-QSAR models capable of accelerating drug discovery in an era of big data and precision medicine.
Keyword :
Artificial intelligence, Deep learning, Quantitative Structure–activity Relationship (QSAR), Drug discovery, Cheminformatics, Predictive modeling, Medicinal chemistry