CNNCaps-DBP: Leveraging protein language models with attention-augmented convolution for DNA-binding protein...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• DNA-binding proteins (DBPs) are critical for DNA replication and gene regulation, playing roles in diseases like cancer.
• Experimental DBP identification is time-consuming and costly.
• Current prediction models often fail to effectively utilize features from pre-trained protein language models.

Purpose of the Study:

• To develop a novel, accurate, and efficient computational method for predicting DNA-binding proteins (DBPs) using only primary sequence information.
• To address limitations in existing DBP prediction models, particularly their feature extraction from pre-trained models.

Main Methods:

• Proposed CNNCaps-DBP, a deep learning model integrating the ESM C pre-trained protein language model.
• Employed an attention-augmented convolution module to enhance protein embeddings.
• Utilized a hybrid Capsule network and MLP architecture for prediction, optimized with a dynamic learning rate scheduler.

Main Results:

• CNNCaps-DBP demonstrated significantly superior predictive performance compared to existing models.
• The model maintained high performance on independent datasets, outperforming state-of-the-art methods.
• Case studies confirmed the model's strong predictive capability for DBP identification.

Conclusions:

• CNNCaps-DBP offers a robust and efficient computational framework for accurate DBP prediction from sequence data.
• The method overcomes limitations of previous approaches by effectively leveraging pre-trained protein language models.
• This advancement facilitates research into protein function and disease mechanisms linked to DBPs.