AI4Molecular Design

Leveraging deep learning approaches for the prediction and design of protein allostery

The laboratories of Pierre Vandergheynst (STI) and Patrick Barth (SV) will be hosting the 5th CIS collaboration grant.

Abstract:

Proteins are not static molecules. They often behave as switches, alternating between states that carry out distinct functions. Hence, switching is a powerful mechanism of biological regulation and typically occurs upon structural changes triggered by a wide variety of external stimuli (e.g. from photon absorption to the binding of another protein) through a process referred as allostery. Understanding how this switching behavior occurs at the molecular level remains challenging. The advent of deep learning offers new opportunities to explore and predict protein motions but these approaches have mostly been applied to static representations of protein structures. Here, we propose to tackle the modeling of switching dynamics using specially designed geometric deep learning techniques applied to molecular representations. We expect that our approach will lead to a new understanding of the mechanisms underpinning these switches and, eventually, to generative approaches allowing to engineer novel molecular switches. 

 

Associated Labs

LTS2 is a team of researchers led by Prof. Pierre Vandergheynst working within the Institute of Electrical Engineering of the EPFL, one of the two Swiss federal institutes of technology. The main part of our research activities focuses on modern challenges in data processing.

The joint expertise of the acoustic group extends the LTS2 research landscape to audio engineering and electroacoustics.

At the LPCE, we work at the interface of biophysics, chemical, structural, computational and cell biology to uncover the molecular principles that regulate protein and cellular signaling. We use this understanding to (1) design protein systems with novel biosensing and signaling functions for synthetic biology and engineered cell therapeutic applications; (2) predict the effects of genetic variations on protein structure/function for personalized cancer medicine applications.

We are particularly interested in deciphering the molecular underpinnings regulating signaling and transport across biological membranes, which control cellular processes but have been particularly challenging to study experimentally.

Opportunities

Postdoctoral Position in Deep Learning for Structural Biology and Protein Design

Contact

For more information, please contact: [email protected]