HIV-1 M group subtype classification using deep learning approach.

Area of Science:

• Bioinformatics
• Genomics
• Machine Learning

Background:

• Traditional HIV-1 M group subtype classification relies on statistical methods limited by sample size.
• Accurate HIV-1 subtype identification is crucial for epidemiological tracking and treatment strategies.

Purpose of the Study:

• To introduce HIV-1-M-SPBEnv, the first deep learning-based method for HIV-1 M group subtype classification.
• To overcome sample size limitations inherent in traditional classification methods.
• To provide a highly accurate and accessible tool for HIV-1 subtype prediction.

Main Methods:

• Development of HIV-1-M-SPBEnv, a deep learning model utilizing a convolutional Autoencoder with residual blocks and a fully connected neural network.
• Generation of a synthetic dataset using artificial molecular evolution to address sample size constraints.
• Validation of the model's performance on an independent dataset.

Main Results:

• HIV-1-M-SPBEnv achieved 100% precision, accuracy, recall, and F1 score in classifying all 12 HIV-1 M group subtypes.
• The model effectively simplifies high-dimensional DNA sequence data into low-dimensional representations.
• Independent dataset validation confirmed the model's robust classification capability.

Conclusions:

• HIV-1-M-SPBEnv offers a significant advancement in HIV-1 M group subtype classification, surpassing traditional methods.
• The deep learning approach effectively handles complex genetic data and overcomes sample size limitations.
• The publicly accessible web server and code empower researchers and clinicians with precise HIV-1 subtype identification tools.