Ligand-based virtual screening (LBVS) is a cornerstone of computational drug discovery, offering a receptor-independent approach to identify bioactive compounds by analyzing physicochemical and structural features of ligands. Traditional LBVS methods, such as molecular similarity searches and quantitative structure-activity relationship (QSAR) modeling, leverage structural fingerprints and pharmacophore models to predict biological activities efficiently. Advances in artificial intelligence (AI) have revolutionized LBVS by introducing machine learning and deep learning techniques. These innovations enhance molecular representation, optimize similarity searches, and refine QSAR models with unprecedented accuracy. AI-powered frameworks now incorporate multi-task learning, transfer learning, and graph-based neural networks to tackle data scarcity and integrate diverse molecular features. Cutting-edge algorithms, such as Chemception and DeepCDA, significantly outperform conventional approaches, enabling large-scale virtual screening with improved predictive power and generalizability. Despite these advancements, challenges remain in quantifying model uncertainty and identifying structurally novel scaffolds. This chapter explores the evolution of LBVS, emphasizing the transformative role of AI in advancing molecular fingerprinting, pharmacophore modeling, and QSAR methodologies. Future directions include the integration of AI with hybrid models and the development of robust, interpretable systems for ligand discovery.