BiCaps-DBP: Predicting DNA-binding proteins from protein sequences using Bi-LSTM and a 1D-capsule network

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• DNA-binding proteins (DBPs) are vital for numerous cellular functions, including DNA replication, transcription, repair, and splicing.
• Identifying DBPs is critical for pharmaceutical research, particularly in human cancers and autoimmune diseases.
• Experimental methods for DBP identification are often time-consuming and expensive, necessitating efficient computational approaches.

Purpose of the Study:

• To develop a rapid and accurate computational method for predicting DNA-binding proteins (DBPs) from primary sequences.
• To introduce BiCaps-DBP, a novel deep learning model designed to enhance DBP prediction performance.

Main Methods:

• The study employed a deep learning approach, combining bidirectional long short-term memory (BiLSTM) with a 1D-capsule network.
• The BiCaps-DBP model was trained and evaluated on three distinct training and independent datasets to assess its generalizability and robustness.

Main Results:

• BiCaps-DBP demonstrated superior performance compared to an existing predictor across three independent datasets.
• The proposed model achieved accuracy improvements of 1.05% on PDB2272, 5.79% on PDB186, and 0.40% on PDB20000.
• These results highlight the model's effectiveness and reliability in DBP prediction.

Conclusions:

• BiCaps-DBP represents a significant advancement in computational DBP prediction.
• The method offers a promising and accurate alternative to traditional experimental techniques.
• This deep learning approach has the potential to accelerate genomic annotation and drug discovery efforts.