Foundation models for low-data drug discovery: mining heterogeneous public and proprietary data in neglected indications.

INTRODUCTION: This review examines how Foundation Models can address critical limitations of data scarcity in drug discovery, particularly for neglected diseases where traditional approaches are ineffective. It highlights the need for new methodologies that integrate heterogeneous data sources to enable more equitable and efficient therapeutic development. AREAS COVERED: This review synthesizes recent advances in Foundation Models and related machine learning approaches for low-data drug discovery, with a focus on applications in neglected diseases. The authors searched PubMed, Scopus, and Web of Science for available literature using terms related to foundation models, machine learning, deep learning, AI with neglected diseases, drug discovery, low-data settings, transfer learning, and related methodological and disease-specific terms. Reference lists of included reviews were additionally screened for relevant primary literature. EXPERT OPINION: Foundation Models are poised to play a central role in drug discovery. Nevertheless, their effectiveness for low-data and neglected diseases will depend on strong collaboration, responsible and ethical use, and continued technical innovation.