Small data, big challenges: Machine- and deep-learning strategies for data-limited drug discovery

A critical bottleneck limiting the potential of Machine Learning (ML) and Deep Learning (DL) models within the drug discovery and development (DDD) pipeline is the scarcity of high-quality experimental data. Limited data is not an anomaly but an inherent characteristic of the DDD process. Significant financial costs, time, and confidentiality concerns limit the scale of available datasets. Applying standard ML and DL algorithms directly to these small datasets presents substantial challenges. Traditional ML models remain constrained by their dependence on handcrafted features and limited ability to capture complex biological relationships. In contrast, DL algorithms that assume data abundance are prone to overfitting and poor generalization when trained on small datasets. The small data problem thus represents a fundamental constraint that shapes the practical utility and trustworthiness of AI applications in DDD. While prior reviews have surveyed the broad landscape of AI and ML in drug discovery, a significant gap exists concerning the small data challenge across the DDD pipeline. Addressing this challenge requires adapting DL methods that typically assume data abundance, while also extending traditional ML approaches that, although well-suited to small data, remain limited in their representational capacity. This review addresses this gap by surveying key drug discovery tasks, highlighting the prevalence of limited data, and synthesizing both traditional ML methods and advanced DL strategies tailored to these contexts. By integrating methodological advances with task-specific applications, the review outlines current approaches and identifies opportunities for advancing robust, interpretable, and generalizable AI in drug discovery.