Protein Secondary Structure Prediction Using Ensemble Neural Networks With Local And Long-range Amino-acid Features
Predicting protein structures from sequences is a challenging problem. Determining the secondary structures of the protein is an effective approach to infer the complete protein structure. The interactions of local and long-range amino-acid residues in proteins are key contributors in defining the p...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://eprints.usm.my/52692/1/FAWAZ%20HAMEED%20HAZZAA%20MAHYOUB%20-%20TESIS24.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Predicting protein structures from sequences is a challenging problem. Determining the secondary structures of the protein is an effective approach to infer the complete protein structure. The interactions of local and long-range amino-acid residues in proteins are key contributors in defining the protein secondary structures. Recent works have focused on capturing local and long-range amino-acid interactions using various predicted protein structural features via an ensemble of deep learning techniques. Nevertheless, determining these structural features is always associated with intensive computing. Moreover, their predictive performance is heavily relied on the quality of the data features resulting from evolutionarily related proteins. This study proposes a method for predicting protein secondary structure by incorporating Feed-Forward Neural Network (FFNN) with bidirectional Long Short-Term Memory (LSTM) networks to capture local and long-range amino-acid interactions. To further improve the prediction accuracy of proteins with few evolutionarily related proteins, additional data features based on the physicochemical properties of amino acids have been proposed. The empirical outcomes indicate that the proposed method in this study shows competitive prediction accuracy compared to Sequence-based Prediction Online Tools for one dimensional structural features (SPOT-1D) and PORTER5. In addition to that, the method outperformed several well-known cutting-edge methods by 2-3 percentagepoint improvement. |
|---|