DeepRHD: An efficient hybrid feature extraction technique for protein remote homology detection using deep learning...

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Protein Remote Homology Detection (PRHD) is crucial for identifying protein structure and function.
• Variable sequence lengths and the vast number of daily generated sequences pose challenges for evolutionary information derivation.
• Existing methods struggle to correlate distantly related protein sequences effectively.

Purpose of the Study:

• To propose an efficient technique for predicting homology in distantly located protein sequences.
• To develop a novel deep neural network model for protein sequence classification.
• To integrate large datasets with extensive feature sets for improved homology detection.

Main Methods:

• A deep neural network, specifically a CNN-GRU model, was employed for protein sequence classification.
• Feature extraction methods were utilized based on different protein families.
• The proposed model, DeepRHD, was trained and tested on the SCOP benchmark dataset.

Main Results:

• The DeepRHD model demonstrated superior performance compared to state-of-the-art methods.
• The model was evaluated on an average of 8000 sequences per superfamily from the SCOP dataset.
• The proposed technique successfully integrates massive datasets with a large feature set.

Conclusions:

• The developed DeepRHD model offers an efficient and accurate approach to Protein Remote Homology Detection.
• This method can aid in the early detection of diseases such as sickle cell anemia and influenza.
• The model provides a foundation for subsequent drug development strategies.