Loading...

Improvement on Deep Learning based Methods of Protein Structure Prediction

Ahmadiyan, Hadis | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57777 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Koohi, Somayye
  7. Abstract:
  8. Proteins are natural polymers composed of amino acids, which, according to their structure, can interact with other molecules and play a wide range of roles. Predicting the tertiary structure of proteins is fundamental for explaining their function and for applications such as drug design. Considering the time-consuming and expensive laboratory methods on one hand and the rapid growth and large volume of protein data on the other hand, a faster solution for finding protein structure in large scales is needed. With the increasing progress of artificial intelligence and neural networks, the use of this approach to predict the three-dimensional structure of proteins using their amino acid sequence has been of great interest. However, these methods still face the problem of prediction accuracy, interpretability, limitations such as the need to search for similar and homologous sequences, multiple alignments, and finally, consuming a lot of computing resources. In this study, focusing on improving speed and computation bottlenecks, employing more optimized tools for search and alignment generation, we have accelerated this stage by a factor of 2 without losing prediction accuracy. We have also tried to take a step towards the interpretability of these models by analyzing the effect of the homologous search tools results on the final prediction accuracy
  9. Keywords:
  10. Deep Learning ; Protein Structure ; Homology ; Interpretability ; Multiple Alignment ; Protein Structure Prediction ; Sequence Search

 Digital Object List

 Bookmark

No TOC