A groundbreaking study published in Health Data Science, a Science Partner Journal, introduces the Knowledge-Empowered Drug Discovery (KEDD) framework, poised to revolutionize drug discovery. KEDD integrates structured and unstructured knowledge, enhancing AI-driven exploration of molecular dynamics and interactions.
Traditionally, AI applications in drug discovery have been limited by their narrow focus, overlooking the wealth of structured and unstructured data. Professor Zaiqing Nie from Tsinghua University's Institute for AI Industry Research highlights KEDD's potential to enhance drug discovery by synergizing data from molecular structures, knowledge graphs, and biomedical literature.
KEDD employs robust representation learning models to distil dense features from diverse data modalities. It integrates these features through fusion processes and predictive networks, demonstrating superiority over existing models. Notably, KEDD addresses the 'missing modality problem' by leveraging sparse attention and modality masking techniques, utilising existing knowledge bases for predictions.
Yizhen Luo, a key contributor, envisions enhancing KEDD with multimodal pre-training strategies, aiming to establish a versatile, knowledge-driven AI ecosystem accelerating biomedical research for therapeutic discovery and development.
Click here to read the original news story.