DeepCR: predicting cytokine receptor proteins through pretrained language models and deep learning networks

Area of Science:

• Computational biology
• Immunology
• Protein science

Background:

• Cytokine receptors are key mediators of immune responses and are implicated in cytokine storms, contributing to diseases like ARDS and autoimmune disorders.
• Accurate identification of cytokine receptors is vital for understanding their functions, developing therapeutic targets, and guiding clinical treatments.
• Traditional methods for identifying cytokine receptors are inefficient, necessitating advanced computational approaches.

Purpose of the Study:

• To develop a novel, efficient, and accurate computational method for classifying cytokine receptor proteins.
• To address the gap in dedicated studies for cytokine receptor classification among membrane proteins.

Main Methods:

• A hybrid framework combining pre-trained language models (PLMs) and a multi-window convolutional neural network (mCNN) was developed.
• PLMs (e.g., ProtTrans, ESM variants) were used to extract biochemical context from raw protein sequences.
• mCNN architecture with varying window sizes was employed to capture local and global sequence patterns.

Main Results:

• The proposed model achieved high performance, with an Area Under the Curve (AUC) of 0.96 in training and 0.97 and 0.93 in independent tests.
• The model demonstrated significant effectiveness in distinguishing cytokine receptors from non-cytokine receptor proteins.
• The method eliminates the need for manual feature extraction, offering a robust and scalable solution.

Conclusions:

• The novel PLM-mCNN framework provides a fast and accurate method for identifying cytokine receptor proteins.
• This approach has significant implications for drug discovery and advancing the understanding of cytokine-mediated diseases.
• The study highlights the potential of machine learning in specialized protein classification tasks.