BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Molecular deep learning at the edge of chemical space.
DTSTART:20250204T151500
DTEND:20250204T161500
DTSTAMP:20260501T234603Z
UID:abc5dc053beb6c56464960fe79ce5cddbe74b46e0c589f72398a6634
CATEGORIES:Conferences - Seminars
DESCRIPTION:Derek van Tilborg\nMolecular machine learning models often fai
 l to generalize beyond the chemical space of their training data\, limitin
 g their ability to reliably perform predictions on structurally novel bioa
 ctive molecules. To advance the ability of machine learning to go beyond t
 he ‘edge’ of their training chemical space\, we introduce a joint mode
 ling approach that combines molecular property prediction with molecular r
 econstruction\, enabling us to estimate model generalizability through a n
 ew reconstruction-based ‘unfamiliarity’ metric. Via a systematic analy
 sis spanning more than 30 bioactivity datasets\, we demonstrate that unfam
 iliarity not only effectively identifies out-of-distribution molecules but
  also serves as a reliable predictor of classifier performance. We show th
 at our method is independent from well-established methods to estimate pre
 diction reliability and that we can successfully identify high-confidence 
 predictions on molecules that are structurally novel – a central challen
 ge in drug discovery. Our findings highlight that joint modelling can be a
  powerful strategy for extending the reach of machine learning models into
  uncharted regions of chemical space\, advancing the discovery of diverse 
 and novel molecules.\n\nDerek van Tilborg is a PhD student in the molecula
 r machine learning team of Francesca Grisoni at the Eindhoven University o
 f Technology (Chemical Biology\, Dept. Biomedical Engineering). Having an 
 academic background in both biomedical sciences and bioinformatics\, he ha
 s developed a passion for artificial intelligence in the realm of drug dis
 covery. Derek works on improving how machine learning approaches are appli
 ed to drug screening data to bridge the gap between computational methods 
 and experiments. His research interests are focussed on solving the practi
 cal bottlenecks that accompany prospective studies\, such as the implement
 ation of active learning and handling out-of-distribution molecular data.
LOCATION:https://epfl.zoom.us/j/68447908297?pwd=OU5JUGJUSUhZc0ZNYjQ2WENvYI
 NRdz09
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR