Back to search

Institutional Repository of Vilnius University

Giliojo mokymosi taikymas baltymų struktūrų analizei /

Abstract

dc:description

This bachelor's thesis is dedicated to exploring the application of deep learning networks in protein structure analysis. Special attention is given to advanced neural network architectures: recurrent neural networks (RNNs), transformer neural networks, AlphaFold protein structure prediction tool and the ESM-1b protein language model. To evaluate AlphaFold3’s ability to address real-world protein folding challenges, we conducted a provocative experiment using proteins in their apo and holo conformations. We examine how tools based on neural network architectures are applied to the structural and functional analysis of Sphagnum divinum proteins. Using the ESM-1b protein language model, we compared and supplemented the results of previous annotation studies on Sphagnum divinum proteins. We also identified and proposed hypotheses regarding the involvement of unannotated proteins in specific cellular processes. The thesis covers the following key aspects: 1. A literature review of the advantages of recurrent neural networks (RNNs) and transformers for amino acid interaction modelling. 2. An evaluation of AlphaFold3’s ability to predict proteins in apo and holo conformations and to reconstruct conformational diversity. A comparison of the conformational diversity reconstruction capabilities of AlphaFold3 and AlphaFold2. 3. A literature review of experiments using AlphaFold2 to annotate the Sphagnum divinum proteome. 4. Identification of hypothetical functions of Sphagnum divinum proteins using the ESM-1b protein language model, and comparison of the obtained results with experimental data. The studies revealed that AlphaFold3 is an effective and accurate tool for protein structure prediction and analysis, and it is being tested for application in organism-wide proteome annotation. However, this model also has limitations, especially regarding the prediction accuracy for proteins from heterogeneous families and the recreation of conformational diversity. Beyond AlphaFold tools, other protein language models based on deep neural network architectures are being developed and applied. Before experimental validation, we used the ESM-1b protein language model to determine hypothetical protein functions and identify protein involvement in certain biological processes. Due to the rapid advancement and application of deep neural networks in biological research, researchers contribute to their improvement and adaptation to address new biological problems. Integrating deep learning methods into bioinformatics may facilitate faster and broader understanding of protein functions and potentially accelerate the process of new biological discoveries.

Degree

thesis:*
Grantor dc:publisher
Institutional Repository of Vilnius University
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Mažuolis, Martynas,

Rights

dc:rights
Statement dc:rights
  • info:eu-repo/semantics/openAccess
Language dc:language
lit

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
Vilnius University
Base URL
epublications.vu.lt/oai
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Mažuolis, Martynas,. Giliojo mokymosi taikymas baltymų struktūrų analizei /. Institutional Repository of Vilnius University, 2025. https://repository.vu.lt/VU:ELABAETD239029558&prefLang=en_US