BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Molecular set representation learning
DTSTART:20231024T151500
DTEND:20231024T161500
DTSTAMP:20260416T053904Z
UID:7f4917f56c82404ce15b126a95b64bbbd7cb883107cd3723883f9083
CATEGORIES:Conferences - Seminars
DESCRIPTION:Daniel received his BSc in computer science at the Bern Univer
 sity of Applied Sciences in 2013 and his MSc in Bioinformatics and Computa
 tional Biology at the University of Bern in 2016. In 2020 he received his 
 PhD in Chemistry and Molecular Sciences for his thesis “Scalable Methods
  for the Exploration and Visualization of Large Chemical Spaces” from th
 e University of Bern under the supervision of  Prof. Jean-Louis Reymond. 
 His main research interest is efficient machine learning and data visualis
 ation applied to natural sciences\, focusing on the intersection of chemis
 try and biology. After a two-year stay as a permanent research staff membe
 r at  IBM Research in the Team of Teodoro Laino working on machine learni
 ng for biocatalysis\, he started as a postdoctoral researcher in the group
  of  Prof. Pierre Vandergheynst at EPFL.\nComputational representation of
  molecules can take many forms\, including graphs\, string-encodings of gr
 aphs\, binary vectors\, or learned embeddings in the form of real-valued v
 ectors. These representations are then used in downstream classification a
 nd regression tasks using a wide range of machine-learning models. However
 \, existing models come with limitations\, such as the requirement for cle
 arly defined chemical bonds\, which often do not represent the true underl
 ying nature of a molecule. Here\, we propose a framework for molecular mac
 hine learning tasks based on set representation learning. We show that lea
 rning on sets of atomic invariants alone reaches the performance of state-
 of-the-art graph-based models on the most-used chemical benchmark data set
 s and that introducing a set representation layer into graph neural networ
 ks can surpass the performance of established methods in the domains of ch
 emistry\, biology\, and material science. We introduce specialised set rep
 resentation-based neural network architectures for reaction yield and prot
 ein-ligand binding affinity prediction. Overall\, we show that the techniq
 ue we denote molecular set representation learning is both an alternative 
 and an extension to graph neural network architectures for machine learnin
 g tasks on molecules\, molecule complexes\, and chemical reactions.
LOCATION:MA A1 10 https://plan.epfl.ch/?room==MA%20A1%2010
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR