DeepHIV: A Sequence-Based Deep Learning Model for Predicting HIV-1 Protease Cleavage Sites

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Human immunodeficiency virus type 1 (HIV-1) drives acquired immunodeficiency syndrome (AIDS).
• Identifying HIV-1 protease cleavage sites (PCSs) is crucial for developing novel anti-AIDS therapeutics.
• Computational prediction of PCSs aids in discovering cleavable substrates and understanding substrate specificity.

Purpose of the Study:

• To develop a deep learning model, DeepHIV, for predicting HIV-1 PCSs solely from substrate sequence information.
• To enhance the accuracy and robustness of HIV-1 PCS prediction.

Main Methods:

• A deep learning model, DeepHIV, was designed.
• It employs a convolutional neural network with an attention mechanism to extract contextual features from amino acid sequences.
• A biased support vector machine classifier addresses the imbalance between cleavable and uncleavable substrates.

Main Results:

• DeepHIV demonstrated superior performance compared to existing state-of-the-art prediction methods.
• The model achieved high accuracy across all benchmark datasets and evaluation metrics.
• DeepHIV effectively leverages sequence information to learn latent substrate features.

Conclusions:

• DeepHIV is an accurate and robust computational tool for predicting HIV-1 PCSs.
• The model's success highlights the capability of deep learning in analyzing sequence data for biological insights.
• This tool can facilitate the design of more effective anti-AIDS inhibitors.