Subject

Deep learning models that predict protein sequence and structure (for instance ProteinMPNN or Alphafold) represent a novel approach in computational biology. Such models are usually composed of graph neural networks. However, the black-box nature of such models undermines its wider applicability, particularly in critical domains such as biology, where interpretability is required. This master thesis seeks to bridge this gap by revising explainability techniques existing in the deep learning literature and apply them into computational biology deep models. The main objective is to provide insights into the decision-making process of ProteinMPNN or Alphafold kind of models.

Kind of work

Objectives:
1. Review and Analyze Existing Explanation Techniques: Conduct an in-depth examination of existing methodologies for generating explanations in deep learning and specifically graph deep learning models (SA, LRP, GradientxInput...)
2. Develop Tailored Explanation Techniques for ProteinMPNN (or any other kind of computational biology model at protein level): Propose a novel approach specifically tailored to the architecture and functionality such model, leveraging the graph-based approach and features representations.
3. Quantitative and Qualitative Evaluation: Establish a comprehensive evaluation framework for biologists, with quantitative metrics and qualitative assessments to study the relevance of the generated explanations.

Framework of the Thesis

Framework of the Thesis
• Robust deep learning based protein sequence design using ProteinMPNN (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9997061/)
• Highly accurate protein structure prediction with AlphaFold (https://www.nature.com/articles/s41586-021-03819-2)
Graph Neural Networks (GNNs):
• ‘Graph Neural Networks: A Review of Methods and Applications” Jie Zhou et al.
• "A Gentle Introduction to Graph Neural Networks", Sanchez-Lengeling, et al., 2021. (https://distill.pub/2021/gnn-intro/)
Explanation techniques for GNNs:
• “Higher-Order Explanations of Graph Neural Networks via Relevant Walks” T. Schnake et al.
• “Explainability Techniques for Graph Convolutional Networks” F. Baldassarre et al 2019.
• “Explainability Methods for Graph Convolutional Neural Networks” E. Pope et al, 2019.
• “GNNExplainer: Generating Explanations for Graph Neural Networks.” Rex Ying et al, 2019
• ‘Explainability in Graph Neural Networks: A Taxonomic Survey’ Hao Yuanet al, 2021.

Number of Students

1 or 2

Expected Student Profile

Good knowledge of Machine Learning, AI and data processing. Good programming skills in Python (particularly PyTorch)