Identification of Antioxidant Proteins With Deep Learning From Sequence Information

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Antioxidant proteins play a vital role in disease control by neutralizing harmful free radicals.
• Accurate identification of antioxidant proteins is increasingly important due to their medicinal value.
• Traditional machine learning methods often struggle with the nonlinear and unbalanced nature of biological data.

Purpose of the Study:

• To develop a robust deep learning-based classifier for identifying antioxidant proteins.
• To leverage advanced feature extraction and dimensionality reduction techniques for improved classification accuracy.
• To provide a user-friendly tool for the scientific community to facilitate antioxidant protein identification.

Main Methods:

• Utilized a deep autoencoder for nonlinear representation learning from raw input data.
• Employed mixed g-gap dipeptide composition as a feature vector.
• Applied t-Distributed Stochastic Neighbor Embedding (t-SNE) for dimensionality reduction.
• Performed Support Vector Machine (SVM) for the final classification task.

Main Results:

• The proposed deep learning classifier achieved a high F1 score of 0.8842.
• The classifier demonstrated a high Matthews Correlation Coefficient (MCC) of 0.7409 in 10-fold cross-validation.
• The method significantly outperformed traditional machine learning approaches in antioxidant protein identification.

Conclusions:

• The developed deep learning classifier is a promising and effective tool for identifying antioxidant proteins.
• The proposed method addresses the limitations of traditional machine learning in handling complex biological data.
• A publicly accessible web server, IDAod, has been created to aid researchers in antioxidant protein identification.