Related Experiment Video
Updated: Aug 6, 2026
08:04
Identification and Classification of Position-specific GABAA Receptor Subunit Missense Variants for Their Role In Hippocampal Pyramidal Neurons
Published on: June 6, 2025
Clinically interpretable deep learning for breast cancer missense variant pathogenicity prediction
Rahaf M Ahmad1, Noura AlDhaheri1, Mohd Saberi Mohamad1,2,3
1Department of Genetics and Genomics, College of Medical and Health Sciences, United Arab Emirates University, Al-Ain, United Arab Emirates.
Frontiers in Bioinformatics
|August 5, 2026
Summary
This study introduces a deep learning framework to accurately predict breast cancer missense variant pathogenicity, addressing a critical need for scalable and interpretable computational tools in cancer genomics.
Area of Science:
- Genomics
- Computational Biology
- Artificial Intelligence in Medicine
Background:
- Missense variants in breast cancer present diagnostic challenges due to functional diversity.
- Current laboratory assays for variant pathogenicity are costly and not scalable.
- There is a need for accurate, scalable, and interpretable computational methods for variant classification.
Purpose of the Study:
- To develop and evaluate a novel deep learning framework for predicting breast cancer missense variant pathogenicity.
- To integrate comprehensive data preprocessing, imputation, and model benchmarking.
- To enhance the transparency and clinical interpretability of variant pathogenicity predictions.
Main Methods:
- Curated genetic variants from multiple databases and annotated using Ensembl Variant Effect Predictor (VEP).
- Employed Variational Autoencoders (VAE) for missing-value imputation.
- Trained and evaluated seven deep learning models (MLP, CNN, DNN, RNN, LSTM, GRU, Transformer) using 11 performance metrics and five random seeds.
- Utilized recursive feature elimination, permutation importance (PMI), and Local Interpretable Model-Agnostic Explanations (LIME) for transparency.
Main Results:
- GRU model achieved the highest internal AUC (0.9956), with strong precision (0.9967) and calibration (ECE 0.0095).
- LSTM model demonstrated superior external performance (AUC 0.9457), outperforming eleven standalone predictors.
- Model predictions showed alignment with conservation signals (phyloP470way, Eigen-PC scores).
- The framework provides confidence intervals and LIME visualizations for enhanced interpretability.
Conclusions:
- This study presents a comprehensive evaluation of deep learning for breast cancer variant classification.
- The proposed framework combines high-performance models with interpretable AI tools for reproducible benchmarking.
- It offers a foundation for future research and clinical translation in cancer genomics.