TREPP: Tandem Repeat Expansion Pathogenicity Prediction via Stacked CatBoost and Context-Aware Sequence Features.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Predicting pathogenic potential of tandem repeat (TR) expansions is crucial for neurogenetic disorder research.
• Existing methods often overlook intrinsic sequence properties that influence TR instability.
• Data scarcity and class imbalance pose significant challenges in TR expansion prediction.

Purpose of the Study:

• To develop an interpretable ensemble learning framework, TREPP, for prioritizing high-risk TR loci.
• To integrate genomic location and local sequence properties for improved prediction accuracy.
• To address data scarcity and class imbalance in TR expansion prediction.

Main Methods:

• TREPP framework combines genomic location and local sequence properties.
• Feature extraction includes gene proximity, GC content differences, and local repeat density.
• A multi-sampling strategy trains multiple CatBoost base learners to mitigate label scarcity.
• A sparse logistic regression meta-learner aggregates base model outputs for a final prioritization score.

Main Results:

• TREPP achieved an AUPRC of 94.71% on a benchmark dataset.
• TREPP outperformed the state-of-the-art method RExPRT by 6.11 percentage points.
• Interpretability and bias-control analyses confirmed TREPP leverages biologically plausible signals.

Conclusions:

• TREPP is an effective and interpretable framework for prioritizing candidate high-risk TR loci.
• The framework's approach addresses key challenges in TR expansion prediction.
• TREPP offers a valuable tool for advancing the understanding of neurogenetic disorders.