OUTRIDER

OUTRIDER detects aberrantly expressed genes from RNA sequencing (RNA-seq) data by modeling gene-wise count co-variation to identify statistically significant outlier read counts indicative of potential pathogenic events in rare disorders.

Key Features:

• Autoencoder-based co-variation modeling: Uses an autoencoder to capture co-variation among genes arising from technical, environmental, or common genetic variation.
• Negative binomial read count model: Models read counts assuming a negative binomial distribution with gene-specific dispersion parameters.
• Expected read count estimation: Estimates expected read counts per gene to compute deviations from expectation.
• Outlier detection: Identifies significant deviations from expected counts to detect aberrant gene expression.
• P-value and FDR adjustment: Performs P-value-based detection with false discovery rate adjustment for multiple testing.
• Co-variation correction with significance thresholds: Combines co-variation correction and significance-based thresholds to avoid arbitrary cutoffs and reduce subjective confounder correction.
• Precision-recall evaluation with simulations: Conducts precision-recall analyses using simulated outlier read counts to evaluate detection accuracy.
• Filtering of non-expressed genes: Provides functionality to filter out non-expressed genes prior to analysis.
• Outlier sample detection: Identifies samples with excessive numbers of aberrantly expressed genes.
• Scalability and computational efficiency: Designed for scalable analysis of large-scale RNA-seq datasets.

Scientific Applications:

• Aberrant expression detection in RNA-seq: Detection of gene-level expression outliers in RNA-seq cohorts.
• Rare disease investigation: Identification of potential pathogenic expression events linked to rare genetic disorders.
• Diagnostic support in genomics: Application in rare disease diagnostic workflows to highlight candidate genes for follow-up.
• Method benchmarking: Evaluation and benchmarking of outlier detection performance using simulated outlier read counts.
• Large-cohort expression analysis: Analysis of cohort-scale RNA-seq data to identify biological or technical outliers.

Methodology:

OUTRIDER applies an autoencoder to model gene co-variation, fits a negative binomial distribution with gene-specific dispersion to estimate expected read counts, tests deviations to produce P-values with false discovery rate adjustment, performs precision-recall analyses using simulated outlier read counts, and includes filtering of non-expressed genes and detection of outlier samples.