Complementary ROC-Derived Indices for Screening Improper Expression Profiles in RNA-Seq Differential Expression...

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Differential expression (DE) analysis of RNA sequencing (RNA-Seq) data is crucial for transcriptomic research.
• Standard DE methods primarily detect monotonic mean shifts, potentially missing genes with non-monotonic expression patterns (improper expression profiles) linked to disease.
• Identifying these improper expression profiles is challenging with current tools.

Purpose of the Study:

• To evaluate receiver operating characteristic (ROC)-based indices, specifically generalized AUC (gAUC) and length of ROC curve (LROC), for screening and prioritizing improper expression profiles in RNA-Seq data.
• To assess the utility of ROC indices as a complement to conventional DE methods for discovering novel disease-associated transcript features.

Main Methods:

• Comparison of DESeq2, classical AUC (cAUC), gAUC, and LROC using simulated negative binomial count data.
• Performance evaluation based on true positive rate and positive predictive value under ranking-based feature selection.
• Application of ROC-derived indices to a public CC miRNA dataset with heuristic screening rules.

Main Results:

• Classical AUC (cAUC) was insensitive to improper expression patterns.
• DESeq2 effectively identified conventionally differentially expressed features but recovered fewer simulated improper profiles.
• gAUC demonstrated superior and stable recovery of improper expression profiles across various simulation scenarios.
• LROC offered complementary insights under low-to-moderate dispersion but was less effective under high dispersion.
• ROC indices identified candidate microRNAs in the CC dataset not prioritized by DESeq2, with some showing biological plausibility.

Conclusions:

• gAUC, with LROC as a supplementary index, offers a practical ROC-based screening extension for RNA-Seq workflows.
• These ROC-derived indices serve as exploratory tools for prioritizing candidate improper expression profiles.
• Independent validation is essential for candidates identified using these heuristic thresholds without controlled error rates.