Protocol-specific and coverage-based RNA-seq metrics characterize RNA integrity signatures across cohorts

Background: RNA degradation profoundly impacts transcript quantification and downstream biological interpretation. Existing RNA quality assessment methods are largely limited to poly(A)-selected mRNA-seq and do not generalize to total RNA-seq, resulting in inconsistent quality evaluations across sequencing protocols. We hypothesize that base-resolution RNA-seq coverage profiles capture protocol-specific quality signatures. By modeling unexpected variation in these coverage patterns, RNA-seq quality can be accurately assessed for both mRNA-seq and total RNA-seq (including frozen and FFPE samples) within a common analytical framework tailored to each protocol. Methods: For both mRNA-seq and total RNA-seq, we quantify abnormal variations in per-base coverage by explicitly modeling degradation-related and other low-quality patterns. For mRNA-seq, specifically, we developed the Degradation Score (DS), which estimates the positional decay in read coverage along transcripts. For total RNA-seq, we introduced the window Coefficient of Variation (wCV), a variant of CV metric that captures coverage nonuniformity, reflecting the tendency of degraded samples to show elevated variability across gene bodies. We applied these metrics to over 2,600 RNA-seq profiles spanning multiple sequencing strategies and cohorts, including TCGA (mRNA-seq), CALGB (fresh frozen total RNA-seq), and ALCHEMIST (FFPE total RNA-seq). Results: Across datasets, our coverage-based metrics showed stronger correlations with conventional QC measures (|r|=0.52-0.59 in mRNA-seq, 0.43-0.89 in total RNA-seq; p<0.001) than the correlations among themselves (|r|=0.24-0.44 in mRNA-seq, 0.44-0.88 in total RNA-seq; p<0.001). We also identified a set of samples in which conventional QC flagged high degradation but coverage profiles appeared intact, and vice versa, highlighting protocol-dependent discrepancies. Manual inspection confirmed markedly greater aberrant variability in samples with high DS (mRNA-seq) or high wCV (total RNA-seq). Using standardized within class sum of squares across RNA quality strata, we found that samples classified as high-quality by our metrics consistently preserved known subtype structure across normalization methods. Conclusions: These protocol-specific, coverage-based measures provide a coherent and reproducible framework for evaluating RNA integrity across sequencing strategies and experimental conditions. Incorporating degradation-aware QC into RNA-seq pipelines improves interpretability, comparability, and biological fidelity of transcriptomic analyses in large cancer genomics studies.