AoP-LSE: Antioxidant Proteins Classification Using Deep Latent Space Encoding of Sequence Features.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Antioxidants are vital for preventing diseases linked to oxidative stress.
• Accurate computational prediction of antioxidants is essential for drug discovery and development.
• Existing methods for antioxidant prediction require separate feature selection and classification models.

Purpose of the Study:

• To develop an effective computational methodology for accurate antioxidant prediction.
• To integrate feature selection and classification into a single deep learning model.
• To enhance the prediction of novel antioxidant compounds.

Main Methods:

• A deep neural network classifier fused with an auto-encoder was employed.
• The model learns class labels within a pruned latent space, integrating feature selection.
• Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (tSNE) visualizations were used for analytical study.

Main Results:

• The proposed method achieved a Matthews Correlation Coefficient (MCC) of 0.43 and a balanced accuracy of 76.2%.
• The model demonstrated superior performance compared to existing computational methods.
• On an independent dataset, the model correctly identified novel antioxidant proteins with 95% accuracy.

Conclusions:

• The integrated deep learning approach effectively harnesses discriminating feature space for improved antioxidant prediction.
• This methodology eliminates the need for separate feature selection and classifier development.
• The proposed computational method shows significant promise for identifying novel antioxidants and advancing disease prevention strategies.