An explainable AI framework integrating machine and deep learning models for multi-species DNA functional group...

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Artificial Intelligence in Biology

Background:

• DNA functional group classification is vital for understanding genetic diversity and evolutionary relationships.
• Machine learning (ML) and deep learning (DL) are increasingly used for DNA sequence analysis, but model interpretability is a challenge.
• Genomic data availability necessitates interpretable methods for biological validation.

Purpose of the Study:

• To present an explainable AI (XAI) framework for multi-species DNA functional group classification.
• To integrate ML and DL models for classifying gene families (e.g., transcription factors, kinases) across Human, Chimpanzee, and Dog datasets.
• To enhance the biological relevance and interpretability of DNA sequence classification models.

Main Methods:

• DNA sequences were converted into k-mers to capture compositional patterns.
• Integrated ML (Logistic Regression) and DL models were trained and hyperparameter-tuned.
• Multi-level XAI techniques (Feature Importance, Saliency Maps, Integrated Gradients, GradientSHAP, Attention Heatmaps) were applied.

Main Results:

• Logistic Regression achieved the highest MCC and F1-scores across all datasets, indicating strong performance.
• Classical models showed better cross-species generalization compared to DL architectures.
• XAI analysis identified consensus and cross-dataset motifs, revealing model stability and fidelity.

Conclusions:

• The developed XAI framework successfully integrates ML and DL for interpretable DNA functional group classification.
• Classical ML models offer a robust and generalizable approach for cross-species genomic analysis.
• XAI methods are crucial for validating biological relevance and understanding model behavior in genomic sequence classification.