DL-TCNN: Deep Learning-based Temporal Convolutional Neural Network for prediction of conformational B-cell epitopes

Area of Science:

• Immunoinformatics
• Computational Biology
• Machine Learning in Medicine

Background:

• Predicting conformational B-cell epitopes (CBCE) is crucial for vaccine design, drug development, and disease diagnosis.
• Traditional laboratory methods are time-consuming and expensive, driving the adoption of computational approaches like Machine Learning (ML).
• Existing ML methods face challenges in achieving high accuracy for CBCE prediction.

Purpose of the Study:

• To introduce a novel Deep Learning-based Temporal Convolutional Neural Network (DL-TCNN) framework for enhanced CBCE prediction.
• To leverage the strengths of deep learning, specifically hybridizing 1D-CNN and TCN architectures, for improved predictive performance.
• To address limitations in existing ML models and improve the accuracy of CBCE prediction.

Main Methods:

• Physicochemical features were extracted from antigen sequences.
• The Synthetic Minority Oversampling Technique (SMOTE) was employed to mitigate class imbalance issues.
• A hybrid DL-TCNN model, combining 1D-CNN and TCN with causal convolutions and dilations, was developed and trained.

Main Results:

• The DL-TCNN model demonstrated high performance on training, validation, and testing datasets.
• Achieved 94.44% accuracy and 0.989 AUC on the training set.
• Reported 85.10% accuracy and 0.855 AUC on the testing set, outperforming existing CBCE prediction methods.

Conclusions:

• The proposed DL-TCNN framework represents a significant advancement in computational CBCE prediction.
• The model's superior performance highlights the potential of deep learning, particularly TCN architectures, in immunoinformatics.
• This approach offers a more accurate and efficient tool for applications in vaccine design and therapeutic development.