Gene driven analytical learning model for accurate breast cancer diagnosis

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning in Oncology

Background:

• Breast cancer prognosis varies significantly due to disease heterogeneity.
• Accurate prognostic prediction is crucial for personalized treatment strategies.
• Gene expression data holds potential for improving breast cancer outcome predictions.

Purpose of the Study:

• To develop an integrated deep learning model for enhanced breast cancer prognosis.
• To identify a robust gene set for prognostic prediction using correlation analysis.
• To validate the model's performance and generalizability across different datasets.

Main Methods:

• Developed a hybrid deep learning model integrating Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks.
• Utilized Pearson correlation to identify a 236-gene set from The Cancer Genome Atlas-Breast Cancer (TCGA-BRCA) data.
• Trained and validated the model on TCGA-BRCA and METABRIC datasets, optimizing with Optuna Bayesian Optimization.

Main Results:

• The hybrid CNN-BiLSTM model significantly outperformed existing machine and deep learning methods.
• Achieved a Recall of 0.9943, ROC AUC of 0.9955, and F1 score of 0.9962, surpassing BiLSTM-only model performance.
• Demonstrated statistical robustness with minimal variance (0.000083) under 20% noise perturbation.

Conclusions:

• The integrated CNN-BiLSTM deep learning framework provides a powerful computational tool for breast cancer precision medicine.
• The model offers highly accurate and consistent prognostic predictions from gene expression data.
• This approach has the potential to improve clinical decision-making and patient outcomes in breast cancer care.