BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
BEGIN:VEVENT
DTSTAMP:20260707T052737Z
UID:27c6b1b1-e141-4801-a911-e6fd546ca5dd
DTSTART:20240507T170000Z
DTEND:20240507T170000Z
DESCRIPTION:This webinar is part of a series run by the ELIXIR 3D-BioInfo C
 ommunity. There is a complete list of webinars here.  \n			\n				 \n			
 \n		\n	\n\n\n\n	\n		\n			\n				The event is hosted by\n\n				\n					\n				
 		\n							\n							\n							\n						\n						\n							\n								Prof. Shos
 hana Wodak (Chair) \n\n								VIB-VUB Center for structural Biology \n	
 						\n							\n								Dr. Gonzalo Parra\n									 \n\n								Barcelo
 na Supercomputing Center  (BSC)  \n							\n							\n								Dr Neela
 dri Sen   \n									 \n\n								University College London (UCL)    
    \n							\n						\n					\n				\n\n				Programme: \n			\n			 \n		\n	
 \n\n\nProstT5: Bilingual Language Model for Protein Sequence and Structure
 \n\nDr. Michael Heizinger (Technical University Munich\, Germany)\n\n\n\nA
 dapting large language models (LLMs) to protein sequences spawned the deve
 lopment of powerful protein language models (pLMs). Concurrently\, AlphaFo
 ld2 broke through in protein structure prediction.\n	Now we can systematic
 ally and comprehensively explore the dual nature of proteins that act and 
 exist as three-dimensional (3D) machines and evolve as linear strings of o
 ne-dimensional (1D) sequences.\n\nHere\, we leverage pLMs to simultaneousl
 y model both modalities by combining 1D sequences with 3D structure in a s
 ingle model. We encode protein structures as token sequences using the 3Di
 -alphabet introduced by the 3D-alignment method Foldseek. This new foundat
 ion pLM extracts the features and patterns of the resulting “structure-s
 equence” representation. Toward this end\, we built a non-redundant data
 set from AlphaFoldDB and fine-tuned an existing pLM (ProtT5) to translate 
 between 3Di and amino acid sequences.\n\nAs a proof-of-concept for our nov
 el approach\, dubbed Protein structure-sequence T5 (ProstT5)\, we showed i
 mproved performance for subsequent prediction tasks\, and for “inverse f
 olding”\, namely the generation of novel protein sequences adopting a gi
 ven structural scaffold (“fold”). Our work showcased the potential of 
 pLMs to tap into the information-rich protein structure revolution fueled 
 by AlphaFold2. ProstT5 paves the way to develop new tools integrating the 
 vast resource of 3D predictions\, and opens new research avenues in the po
 st-AlphaFold2 era. Our model is freely available for all.\n\n\n	Protein Em
 beddings Predict Binding Residues in Disordered Regions’\n\n	Celine Marq
 uet\n		(PhD Student\, Technical University Munich\, Germany)\n\n	The ident
 ification of protein binding residues helps to understand their biological
  processes as protein function is often defined through ligand binding\, s
 uch as to other proteins\, small molecules\, ions\, or nucleotides. Today
 ’s methods predicting binding residues often err for intrinsically disor
 dered proteins or regions (IDPs/IDPRs).\n\n	Here\, we presented a novel ma
 chine learning (ML) model trained to predict binding regions specifically 
 in IDPRs. The proposed model\, IDBindT5\, leveraged embeddings from the pr
 otein language model (pLM) ProtT5 to reach a balanced accuracy of 57.2±3.
 6% (95% confidence interval). This was numerically slightly higher than th
 e performance of the state-of-the-art (SOTA) methods ANCHOR2 (52.4±2.7%) 
 and DeepDISOBind (56.9±5.6%) that rely on expert-crafted features and/or 
 evolutionary information from multiple sequence alignments (MSAs).\n\n	IDB
 indT5’s SOTA predictions are much faster than other methods\, easily ena
 bling full-proteome analyses. Our findings emphasize the potential of pLMs
  as a promising approach for exploring and predicting features of disorder
 ed proteins. The model and a comprehensive manual are publicly available 
 here. \n\n	You can find previous webinars from the 3D-BioInfo Community o
 n the Community webinars page.\n\n	 
SUMMARY:3D-BioInfo: Protein Language Models\, Design and Disorder
URL;VALUE=URI:https://www.elixir-europe.org/events/3d-bioinfo-protein-langu
 age-models-design-and-disorder
END:VEVENT
END:VCALENDAR