FrustraNet: Geometric Deep Learning for Protein Function Prediction through Energetic Frustration Analysis

Mentored by Jens Meiler, Guido Montúfar
at Leipzig University

Protein function emerges from complex energetic landscapes where local regions of high energetic frustration are crucial for functional activities and conformational changes. Furthermore recent studies have identified frustration as a key factor in the dynamics of viral proteins, including SARS-CoV-2. Alas, current protein design methodologies, like ProteinMPNN, have been shown to completely omit frustration patterns in their workflow. As a first step in this direction, the Meiler lab has recently developed FrustraMPNN, which has preliminarily demonstrated the potential of message-passing neural networks in predicting these patterns.

In this project, we propose to develop frustration-guided protein design methodologies through neural networks that predict protein frustration. This work combines geometric deep learning (Montufar) with computational protein structure analysis (Meiler) to create neural network architectures and inference algorithms that enable frustration-aware protein design.

The PhD candidate will work on these research objectives:

1. Develop neural network architectures that efficiently predict protein frustration patterns for integration into protein design pipelines.
2. Create frustration-guided protein design methodologies that preserve functional regions during optimization.
3. Implement these methods specifically for viral proteins that undergo critical conformational changes in SARS-CoV-2 and other pathogens.
4. Validate designed viral proteins experimentally in collaboration with wet-lab researchers.

This interdisciplinary project leverages Prof. Montufar's expertise in developing MPNN architectures with inductive biases specifically designed to capture the mathematical foundations of frustration calculations. His knowledge in geometric deep learning will enable neural networks that respect the physical constraints governing protein frustration. Prof. Meiler's expertise will guide the application of these models to protein engineering and antiviral development.

Building upon foundations established in Prof. Meiler's lab on FrustraMPNN, this project will bring frustration prediction into a design methodology for viral proteins. The approach has potential to advance both viral mechanism understanding and protein design by addressing the stability-function tradeoff in current methods.

Work Environment

The PhD candidate will be based at SECAI, benefiting from collaboration between University Leipzig and TU Dresden, with access to the Max Planck Institute for Mathematics in the Sciences. At the Institute for Drug Discovery, the candidate will apply theoretical advances to protein engineering problems. Machine-learning resources will be available through ScaDS.AI Center, including state-of-the-art CPU and GPU resources.

SECAI offers a first-class environment for advancing your career. You can work with internationally renowned researchers and benefit from the school’s strong networks in industry and research. The graduation of highly qualified researchers is a central project goal in SECAI and doctoral students receive strong support for their professional and personal development.

Prerequisites

• Master's Degree (or equivalent) in bioinformatics, computational biology, mathematics, or related field
• Background in structural biology or protein biophysics
• Programming skills in Python, interest in data analysis and artificial intelligence
• Excellent communication skills and ability to work in interdisciplinary teams