Samtools Benchmarks

RustQC produces samtools-compatible output files (flagstat, idxstats, stats SN section) as part of its single-pass BAM processing. This page compares the output of each tool against the originals on a 10 GB paired-end BAM file (GM12878, ENCODE).

Performance

All samtools-compatible outputs run together in the same single-pass BAM reading that also produces dupRadar, featureCounts, RSeQC, preseq, TIN, and gene body coverage output.

Tool	Traditional	RustQC
samtools flagstat	~1m	—
samtools idxstats	~1s	—
samtools stats	~5m	—
All outputs (single pass)	—	5m 13s

flagstat

Result: Identical

All 16 flagstat metrics match exactly between samtools flagstat and RustQC:

Metric	samtools	RustQC
Total reads	185,718,543	185,718,543
Primary	175,097,721	175,097,721
Secondary	10,620,822	10,620,822
Supplementary	0	0
Duplicates	133,912,519	133,912,519
Primary duplicates	133,912,519	133,912,519
Mapped	185,718,543 (100.00%)	185,718,543 (100.00%)
Primary mapped	175,097,721 (100.00%)	175,097,721 (100.00%)
Paired in sequencing	175,097,721	175,097,721
Read 1	87,559,392	87,559,392
Read 2	87,538,329	87,538,329
Properly paired	174,840,354 (99.85%)	174,840,354 (99.85%)
With itself and mate mapped	174,840,354	174,840,354
Singletons	257,367 (0.15%)	257,367 (0.15%)
Mate on different chr	0	0
Mate on different chr (mapQ>=5)	0	0

The output format is fully compatible with MultiQC and other tools that parse samtools flagstat output.

idxstats

Result: Identical

Per-chromosome read counts match exactly across all 25 reference sequences. Both files include the same reference names, lengths, mapped counts, and unmapped counts, plus the * row for unplaced reads.

stats (Summary Numbers)

Result: Core counts identical, derived metrics have minor differences

RustQC produces the SN (Summary Numbers) section of samtools stats, which is the section parsed by MultiQC for key alignment statistics.

Exact matches (26 of 33 fields)

All primary count fields match exactly:

Metric	Value
raw total sequences	175,097,721
filtered sequences	0
sequences	175,097,721
1st fragments	87,559,392
last fragments	87,538,329
reads mapped	175,097,721
reads mapped and paired	174,840,354
reads properly paired	174,840,354
reads paired	175,097,721
reads duplicated	133,912,519
reads MQ0	345,072
non-primary alignments	10,620,822
total length	17,319,219,330
bases mapped	17,319,219,330
bases duplicated	13,211,215,206
mismatches	53,181,882

Minor differences

Metric	samtools	RustQC	Notes
bases mapped (cigar)	17,220,065,351	17,218,285,734	0.01% difference in CIGAR base counting
average length	99	98	Rounding (true average ~98.9)
average quality	36.4	36.3	Rounding difference
insert size average	1553.3	3084.6	Different outlier filtering strategy
insert size stdev	2276.6	15028.6	Different outlier filtering strategy

The insert size differences are due to samtools applying iterative outlier filtering (removing pairs beyond mean +/- N*stdev, then recomputing), while RustQC uses all properly paired reads. Both approaches are valid; the insert size statistics are informational rather than used for downstream analysis.

The output format includes the samtools stats header comment required for MultiQC parsing.

Benchmark conditions

• BAM file: GM12878 ENCODE RNA-Seq, paired-end, ~185M reads, 10 GB
• samtools version: 1.23 (via Docker)
• RSeQC version: 5.0.4 (via Docker)
• RustQC: 10 threads, --gtf mode
• Hardware: Apple M3 Max, 128 GB RAM