A deep learning model for epidermal growth factor receptor prediction using ensemble residual convolutional neural...

Area of Science:

• Biotechnology
• Bioinformatics
• Computational Biology

Background:

• Epidermal Growth Factor Receptor (EGFR) overexpression drives breast cancer, necessitating effective identification methods.
• Current methods like motif-based approaches and immunohistochemistry have limitations in accuracy, cost, and variability.
• Accurate EGFR identification is crucial for targeted therapies and personalized breast cancer treatment.

Purpose of the Study:

• To develop a novel deep learning predictor, ERCNN-EGFR, for accurate EGFR identification directly from primary amino acid sequences.
• To evaluate and compare various deep learning frameworks for protein feature prediction.
• To establish a cost-effective and scalable computational tool for EGFR detection.

Main Methods:

• Feature extraction using Composition Distribution Transition (CDT), Amphiphilic Pseudo Amino Acid Composition (AmpPseAAC), k-spaced conjoint triad descriptor (KSCTD), and ProtBERT-BFD embeddings.
• Feature refinement via XGBoost-Feature Forward Selection (XGBoost-FFS) to enhance discriminative power.
• Evaluation of deep learning models including BiLSTM, GRU, GAN, and ERCNN, with ERCNN demonstrating superior performance.

Main Results:

• The ERCNN-EGFR model achieved high performance metrics: 93.48% accuracy, 94.53% sensitivity, 92.58% specificity, and a Matthews correlation coefficient of 0.816 post-feature selection.
• The model demonstrated robust performance on an independent test set, achieving 82.85% accuracy.
• Ablation studies identified dual residual building blocks and ProtBERT-BFD features as critical components for the model's predictive accuracy.

Conclusions:

• ERCNN-EGFR provides a scalable, cost-effective, and accurate computational method for identifying EGFR proteins.
• The developed predictor has significant potential applications in breast cancer diagnostics and personalized medicine.
• This approach facilitates efficient therapeutic target discovery for EGFR-driven cancers.