Institutional Repository of Vilnius University
Protein thermostability prediction using sequence representations from protein language models /
Abstract
dc:descriptionProteins, which are stable at higher temperatures, are important for biotechnological applications. Determination of protein's thermostability can be done experimentally, but it might be especially time and resources-demanding, thus it is required to have fast and reliable methods to determine protein's thermal stability. The main objective of this work is to present a method with this functionality. Protein's sequence and structural properties influence thermal stability of the macromolecule. Furthermore, there are transformer architecture-based language models trained to predict probabilities of elements in sequences. Sequences' embeddings - the real numbered vectors representing semantic connections of language components - are created during the unsupervised training. Embeddings can be transferred as input to other models to solve the required tasks using the supervised learning. The main idea of this work is a collection of feed-forward neural networks, which take embeddings from protein language models as input and predict at what temperature the protein is still considered as stable, was trained using embeddings generated for proteins from organisms with annotated growth temperatures. The result of this work is TemStaPro (Temperatures of Stability for Proteins) - a program that demonstrates that embeddings can be efficiently used for protein thermostability prediction task and achieves high accuracy results.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Pudžiuvelytė, Ieva,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD192845392&prefLang=en_US