DeepRHD: An efficient hybrid feature extraction technique for protein remote homology detection using deep learning strategies. (October 2022)

• Record Type: Journal Article
• Authors: Routray, Mukti; Vipsita, Swati; Sundaray, Amrita; Kulkarni, Srinidhi
• Is Part Of: Computational biology and chemistry. Volume 100(2022)
• Publication Date: 2022-10
• Subjects: Protein remote homology -- CNN-GRU -- Deep learning; Chemistry -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; Biochemistry -- Data processing; Biology -- Data processing; Molecular biology -- Data processing; Periodicals; Electronic journals; 542.85
• DOI: 10.1016/j.compbiolchem.2022.107749
• Languages: English
• ISSNs: 1476-9271

Abstract

Abstract: In computational biology, the Protein Remote homology Detection technique (PRHD) has got undeniable significance. It is mostly important for structure and function identification of a protein sequence. The previous years have seen a challenge that lacks postulating a correlation among the sequences. However, the sequences are of variable length. Thereby, it inhibits the proper derivation of evolutionary information among the sequences. The challenges are the usage of physico-chemical properties as a source to get the evolutionary information and the number of sequences generated every day. This however facilitates a new technique to integrate huge amount of data with a massive feature set. In this article, a new and efficient technique is proposed to predict homology for distantly located sequences of proteins. Deep neural network(CNN-GRU model) is used for the classification of the protein sequences. This is based on different protein families and methods of feature extraction.The efficiency of the proposed model DeepRHD is tested on average 8000 sequences per superfamily taken from SCOP benchmark dataset and the results shows that the proposed model is better than other state of art methods. This model is useful in detecting diseases like sickle cell anemia and influenza and developing a drug thereafter. Graphical Abstract: ga1 Highlights: Deep learning model (DeepRHD) to predict homology for distantly located protein sequences. CNN-GRU Model. Trained and tested on SCOP and CATH dataset. Performance comparison with Prodec-BLSTM results. This model outperforms Prodec-BLSTM results.