dupRadar Benchmark

RustQC produces output identical to the R dupRadar package. Every value in the duplication matrix, every count, and the fitted model parameters match exactly. The benchmarks below quantify the performance difference on real RNA-seq data.

Small benchmark

Input: A test BAM file with a chr6-only GTF annotation (2,905 genes).

Metric	dupRadar (R)	RustQC
Runtime	2.50s	0.25s
Speedup	—	10x

Results comparison (small)

Metric	dupRadar (R)	RustQC	Exact match
Intercept	0.03186	0.03186	Yes
Slope	1.60189	1.60189	Yes
Genes total	2,905	2,905	Yes
Genes with reads	636	636	Yes
Genes with duplicates	201	201	Yes
allCounts (unique)	20,449	20,449	100%
filteredCounts (unique)	17,879	17,879	100%
allCountsMulti	22,812	22,812	100%
filteredCountsMulti	20,034	20,034	100%
Total values compared	37,765	37,765	—
Value mismatches	—	0	—

Large benchmark

Input: GM12878 REP1 — a full-size RNA-seq BAM (~10 GB) from the nf-core/rnaseq pipeline, duplicate-marked with Picard. Paired-end, unstranded, aligned to GRCh38 (63,086 genes).

Performance

Tool	Runtime	Max RSS
R dupRadar	27m 21s	N/A (Docker)
RustQC (10 threads)	3m 56s	6.1 GB

Note: RustQC now runs dupRadar, featureCounts, and all 7 RSeQC tools in a single pass, so its runtime reflects the full combined workload. R dupRadar timings are from Docker with x86 emulation on ARM Mac.

Results comparison (large)

Metric	dupRadar (R)	RustQC	Exact match
Intercept	0.8245	0.8245	Yes
Slope	1.6774	1.6774	Yes
Genes total	63,086	63,086	Yes
Genes with reads (unique)	23,597	23,597	Yes
Genes with reads (multi)	24,719	24,719	Yes
allCounts (unique)	14,654,579	14,654,579	100%
filteredCounts (unique)	3,599,832	3,599,832	100%
allCountsMulti	16,089,488	16,089,488	100%
filteredCountsMulti	4,503,920	4,503,920	100%
Total values compared	820,118	820,118	—
Value mismatches	—	0	—

All four count columns match exactly across all 63,086 genes for both unique and multi-mapper counts. A cell-by-cell comparison of the full duplication matrix (820,118 values) shows zero mismatches at a relative tolerance of 1e-6. Model fit parameters (intercept and slope) match to at least 10 significant digits.