Loading...
Search
Search in this resource
sort by
Send to Friend
 Digital Object List
 Bookmark
 Review List
  1. بهبود تفسیرپذیری مدل های پیش بینی کاربرد پروتئین از دنباله پروتئین با استفاده از روش های یادگیری عمیق با قابلیت پیاده سازی نوری
بهبود تفسیرپذیری مدل های پیش بینی کاربرد پروتئین از دنباله پروتئین با استفاده از روش های یادگیری عمیق با قابلیت پیاده سازی نوری
10 viewed

بهبود تفسیرپذیری مدل های پیش بینی کاربرد پروتئین از دنباله پروتئین با استفاده از روش های یادگیری عمیق با قابلیت پیاده سازی نوری

مقیمی اسفندآبادی، وحید الدین Moghimi Esfandabadi, Vahidoddin

Improving Interpretability of Protein Function Predicting Model from Protein Sequence using Deep Learning Methods with Optical Implementation Capability

Moghimi Esfandabadi, Vahidoddin | 2025

10 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58692 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Koohi, Somayyeh
  7. Abstract:
  8. Predicting protein function from amino acid sequences is a fundamental and challenging problem in computational bioinformatics. Despite the remarkable success of deep learning models in this domain, their “black-box” nature and lack of transparency challenge the trustworthiness of their predictions and hinder their use as tools for biological discovery. In response to this dual challenge, this research presents an integrated solution to simultaneously improve prediction accuracy and develop a quantitative, evaluable interpretability method. This solution comprises a model that leverages advanced sequence representations based on large protein language models to extract key features through a convolutional neural network. Furthermore, a framework for quantitative interpretability is employed, which provides a more precise estimation of interpretability accuracy. The model’s interpretation is achieved by computing a global, gradient-based importance vector and subsequently mapping it onto individual amino acids via similarity measurement to identify and evaluate influential regions and motifs within the sequence. Additionally, we investigate the feasibility of hardware implementation of the proposed model on optical platforms. For model evaluation, a standard time-delay protocol, similar to the CAFA challenge, was used with an updated dataset from Swiss-Prot and the Gene Ontology. Experimental results demonstrate the superior performance of the proposed model compared to baseline methods, as well as the high capability of the interpretability method in accurately identifying functional regions. In summary, this research provides a comprehensive solution that not only enhances the accuracy of protein function prediction but also, by introducing an interpretability method, takes a significant step toward transforming deep learning models into reliable and effective tools for scientific
    discovery
  9. Keywords:
  10. Protein Language Model ; Deep Learning ; Gene Ontology ; Convolutional Neural Network ; Protein Function Prediction ; Interpretability ; Optical Neural Networks

 Digital Object List

 Bookmark

...see more