AI foundation models, trained on large-scale data, offer unprecedented opportunities for a wide range of applications. The potentials of these models are further magnified when combined with prompt-based learning, achieving state-of-the-art (SOTA) results even with minimal labeled data.
This presentation delves into the biomedical utilization of three such models: ChatGPT, Segment Anything Model (SAM), and Evolutionary Scale Modeling (ESM). Specifically, we utilized SAM to identify key attributes of pathway entities and their relationships from the figures of research papers. We queried ChatGPT to identify gene relationships by designing effective prompts. To enhance the accuracy of ChatGPT's feedback, we introduced an innovative iterative prompt refinement technique. This method assesses prompt efficacy using metrics like F1 score, precision, and recall. Based on these evaluations, ChatGPT was re-engaged to suggest improved prompts. We also integrated prompt-based learning with SAM to detect proteins in cryo-Electron Microscopy (cryo-EM) images. Furthermore, we prompted the ESM model for signal peptide predictions and large models of single-cell RNA-seq data for data analyses. The outcomes of our studies underscore the potential utilities of foundation models and prompt-based learning for efficient biomedical data analyses and predictions.
Presented by Dr. Dong Xu, Curators' Distinguished Professor in the Department of Electrical Engineering and Computer Science at the University of Missouri-Columbia.
Join Zoom Meeting
Meeting ID: 967 9628 4909