ABEILLE

ABEILLE identifies aberrant gene expression (AGE) from RNA-seq data using variational autoencoders (VAEs) and a decision-tree classifier to detect deviations from learned expression distributions and prioritize candidate pathogenic expression events.

Key Features:

• Variational Autoencoder-based modeling: Employs a variational autoencoder (VAE) to capture complex, high-dimensional RNA-seq gene expression structure without assuming a specific data distribution.
• Single-sample operation: Detects AGEs without requiring biological replicates or control groups, enabling analysis of single-sample clinical datasets.
• Decision Tree Integration: Uses a decision tree to classify genes as aberrantly expressed or non-aberrant based on deviation metrics.
• Anomaly Scoring System: Computes an anomaly score per gene to quantify deviation severity and enable prioritization of AGEs.

Scientific Applications:

• Rare disease diagnosis and research: Identifies aberrantly expressed genes to propose candidate genes involved in rare diseases.
• Complement to genetic testing: Provides expression-based candidate prioritization that complements traditional genetic testing methods.
• Clinical single-sample analysis: Enables detection of pathogenic expression events in settings where replicates or controls are unavailable.

Methodology:

Train a VAE on RNA-seq data to learn the distribution of normal gene expression, detect deviations in new datasets, classify genes with a decision tree based on deviation scores, and compute an anomaly score for each gene.