Specific Aims Page for Student Summer Project – Molcano: In Chemoinformatic Transfer Learning, How Close Is Close Enough?

Zoom information will be sent to e-HAIL members.

A broad goal in drug discovery is to develop chemical foundation models that integrate physical, chemical, and biological evidence for molecular structure, interactions, and functional activity into large-scale, multipurpose deep-learning models. A critical challenge is to assemble these diverse data and identify what data is useful for reliable prediction across diverse tasks. To address this goal, with e-HAIL support, this past summer Haneul Park has developed a repository for ML ready chemical bioactivity datasets from diverse sources. To begin to characterize their complementary value of different datasets, she has clustered them by their chemical similarity and assessed how effective similar datasets are for neural network transfer learning. In this talk, we will share the status of the project and describe preliminary results from this study.

This event will be moderated by Geoffrey Siwo, MD.