Verification of Residue Mutation Free Energy Changes Using Graph Neural Networks
2025 (English)Independent thesis Advanced level (professional degree), 20 credits / 30 HE credits
Student thesis
Abstract [en]
Accurately predicting the effect of single point mutations on proteins is an important task in the fields of biomedical research and bioengineering. One such tool to predict the change in thermodynamic stability is QresFEP-2, a newly developed free energy perturbation method using a hybrid topology protocol. Arguably as important is accurately estimating the error of such tools, as it allows for informed decision making in downstream analysis. In this work, a graph neural network is developed to quantify the errors in QresFEP-2 predictions. Starting from protein structures, proteins are converted to graphs and used by the graph neural network to predict the change in thermodynamic stability. That prediction is then used to estimate the QresFEP-2 error. The error is estimated within 2 kcal/mol in 96% of the cases, and within 1 kcal/mol in 76% of the cases. This shows promising results for future work, both in terms of protein stability prediction and error estimation. It also shows the capability of graph neural networks to learn useful representations from complex data such as protein structures, and how different methods complement each other allowing for more robust predictions.
Place, publisher, year, edition, pages
2025.
Series
UPTEC X ; 25039
Keywords [en]
Graph Neural Network, AI, Machine Learning, Protein stability, enzyme design, error prediction, free energy calculation
National Category
Bioinformatics and Computational Biology Cell and Molecular Biology Structural Biology Computer and Information Sciences
Identifiers
URN: urn:nbn:se:uu:diva-562809OAI: oai:DiVA.org:uu-562809DiVA, id: diva2:1979828
Educational program
Molecular Biotechnology Engineering Programme
Supervisors
Examiners
2025-07-022025-07-012025-07-02Bibliographically approved