Bio Bam Statistics Agent Skill

Version Compatibility

Reference examples tested with: pysam 0.22+, samtools 1.19+

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show <package> then help(module.function) to check signatures
• CLI: <tool> --version then <tool> --help to confirm flags

If code throws ImportError, AttributeError, or TypeError, introspect the installed

package and adapt the example to match the actual API rather than retrying.

BAM Statistics

"Get alignment statistics and coverage from my BAM file" → Generate read counts, mapping rates, per-chromosome statistics, depth profiles, and coverage summaries.

• CLI: samtools flagstat, samtools stats, samtools depth, samtools coverage (samtools)
• Python: pysam.AlignmentFile with pileup() and get_index_statistics() (pysam)

Generate alignment statistics using samtools and pysam.

Quick Summary Commands

| Question | Best tool | Why |

|----------|-----------|-----|

| Quick read counts by FLAG category | samtools flagstat | Fast; counts secondary+supp in totals |

| Per-chromosome counts | samtools idxstats | Fast (needs index); counts secondary+supp |

| Insert size, MAPQ, error, GC | samtools stats -r ref.fa | Comprehensive; feeds MultiQC |

| Per-position depth (small region) | samtools depth or pysam pileup | Slow on full genome |

| Per-position depth (genome-wide) | mosdepth | 3-10x faster than samtools depth |

| Per-region coverage (BED) | mosdepth --by regions.bed | Production default |

| Coverage histogram / cumulative | mosdepth -t 4 --no-per-base | Single-pass histogram |

| Breadth at depth thresholds | mosdepth --thresholds 1,10,30,100 | Standard exome QC |

| Targeted enrichment QC | picard CollectHsMetrics | PCT_OFF_BAIT, FOLD_80_BASE_PENALTY, AT/GC dropout |

| Cross-sample contamination | verifybamid2, somalier | FREEMIX < 0.01 expected |

What Each Tool Counts (and Doesn't)

| Counting category | flagstat | stats | idxstats |

|-------------------|----------|-------|----------|

| Primary alignments | in total minus supp | raw total sequences | mapped column |

| Secondary | secondary line | filtered out | counted in mapped |

| Supplementary | supplementary line | filtered out | counted in mapped |

| Mapping rate denominator | total *including* supp | primary only | mapped+unmapped |

For long-read data where one read produces many supplementary alignments, the senior cross-check:

input_read_count = flagstat_total - secondary - supplementary
                 = stats_raw_total_sequences

Reports of "the file has 1.2M reads" where the input was actually 800k with 400k supplementary chimeric splits trace to flagstat misinterpretation.

samtools flagstat

Fast summary of alignment flags.

samtools flagstat input.bam

Output:

10000000 + 0 in total (QC-passed reads + QC-failed reads)
0 + 0 secondary
50000 + 0 supplementary
0 + 0 duplicates
9800000 + 0 mapped (98.00% : N/A)
9950000 + 0 paired in sequencing
4975000 + 0 read1
4975000 + 0 read2
9700000 + 0 properly paired (97.49% : N/A)
9750000 + 0 with itself and mate mapped
100000 + 0 singletons (1.01% : N/A)
25000 + 0 with mate mapped to a different chr
10000 + 0 with mate mapped to a different chr (mapQ>=5)

Multi-threaded

samtools flagstat -@ 4 input.bam

Output to File

samtools flagstat input.bam > flagstat.txt

samtools idxstats

Per-chromosome read counts (requires index).

samtools idxstats input.bam

Output format: chrom length mapped unmapped

chr1    248956422    5000000    1000
chr2    242193529    4800000    800
chrM    16569        50000      100
*       0            0          150000

Parse idxstats

# Total mapped reads
samtools idxstats input.bam | awk '{sum += $3} END {print sum}'

# Mitochondrial percentage
samtools idxstats input.bam | awk '
    /^chrM/ {mt = $3}
    {total += $3}
    END {print mt/total*100 "% mitochondrial"}'

samtools stats

Comprehensive statistics including insert size, base quality, and more.

samtools stats input.bam > stats.txt

View Summary Numbers

samtools stats input.bam | grep "^SN"

Key summary fields:

• raw total sequences - Total reads
• reads mapped - Mapped reads
• reads mapped and paired - Properly paired
• insert size average - Mean insert size
• insert size standard deviation - Insert size spread
• average length - Mean read length
• error rate - Mismatch rate

Generate Plots (with plot-bamstats)

samtools stats input.bam > stats.txt
plot-bamstats -p plots/ stats.txt

Stats for Specific Region

samtools stats input.bam chr1:1000000-2000000 > region_stats.txt

samtools depth

Per-position read depth.

Basic Depth

samtools depth input.bam > depth.txt

Output: chrom position depth

Depth at Specific Positions

samtools depth -r chr1:1000-2000 input.bam

Include Zero-Depth Positions

samtools depth -a input.bam > depth_with_zeros.txt

Maximum Depth Cap (Critical Trap)

# samtools mpileup historically capped depth at 8000 per position -- the cap was in mpileup, not depth.
# samtools depth -d/--max-depth is deprecated in 1.13+ (silently ignored).
# For mpileup, raise the cap explicitly when working with deep targeted/amplicon data:
samtools mpileup -d 1000000 -f ref.fa input.bam

Pipelines that historically break the 8000 mpileup cap: targeted oncology hotspots (5000-50000x), mitochondrial DNA (small genome, large read share), amplicon viral (ARTIC: 1000-100000x per amplicon), UMI-deduped capture (14000-17000x post-collapse), highly expressed transcripts (rRNA, mt-RNA).

Overlapping Pair Correction

# When fragment length < 2 * read_length, R1 and R2 overlap.
# Default samtools depth double-counts overlap; -s deducts:
samtools depth -s input.bam

Without -s, doubled support inflates somatic VAFs at sites covered by overlapping pairs (especially in fragmented samples: FFPE, cfDNA). mosdepth does not double-count overlap. samtools mpileup and bcftools mpileup both enable overlap detection by default; pass -x/--ignore-overlaps to disable.

mosdepth (Modern Default)

mosdepth -t 4 sample input.bam                                                       # genome-wide per-base
mosdepth -t 4 --by exome.bed --thresholds 1,10,20,30,100 --no-per-base sample input.bam   # exome QC
mosdepth -t 4 --quantize 0:1:10:100: sample input.bam                                # CNV-style bands
mosdepth -t 4 -f ref.fa sample input.cram                                            # CRAM with reference

mosdepth filters dup/secondary/supp by default; configurable via --flag. Memory ~ 4 bytes x longest chrom (1 GB for human chr1, 12+ GB for axolotl). Does not honor base quality; use samtools depth -q INT if needed.

Depth from BED Regions

samtools depth -b regions.bed input.bam

Calculate Mean Depth

samtools depth input.bam | awk '{sum += $3; n++} END {print sum/n}'

samtools coverage

Per-chromosome or per-region coverage statistics (faster than depth).

samtools coverage input.bam

Output columns:

• #rname - Reference name
• startpos - Start position
• endpos - End position
• numreads - Number of reads
• covbases - Bases with coverage
• coverage - Percentage of bases covered
• meandepth - Mean depth
• meanbaseq - Mean base quality
• meanmapq - Mean mapping quality

Coverage for Specific Region

samtools coverage -r chr1:1000000-2000000 input.bam

Coverage from BED

samtools coverage -b regions.bed input.bam

Histogram Output

samtools coverage -m input.bam

pysam Python Alternative

Count Reads

import pysam

with pysam.AlignmentFile('input.bam', 'rb') as bam:
    total = mapped = paired = proper = 0
    for read in bam:
        total += 1
        if not read.is_unmapped:
            mapped += 1
        if read.is_paired:
            paired += 1
        if read.is_proper_pair:
            proper += 1

    print(f'Total: {total}')
    print(f'Mapped: {mapped} ({mapped/total*100:.1f}%)')
    print(f'Properly paired: {proper} ({proper/paired*100:.1f}%)')

Per-Chromosome Counts

import pysam

with pysam.AlignmentFile('input.bam', 'rb') as bam:
    for stat in bam.get_index_statistics():
        print(f'{stat.contig}: {stat.mapped} mapped, {stat.unmapped} unmapped')

Calculate Depth at Position

import pysam

with pysam.AlignmentFile('input.bam', 'rb') as bam:
    for pileup in bam.pileup('chr1', 1000000, 1000001):
        print(f'Position {pileup.pos}: depth {pileup.n}')

Mean Depth in Region

import pysam

def mean_depth(bam_path, chrom, start, end):
    depths = []
    with pysam.AlignmentFile(bam_path, 'rb') as bam:
        for pileup in bam.pileup(chrom, start, end, truncate=True):
            depths.append(pileup.n)

    if depths:
        return sum(depths) / len(depths)
    return 0

depth = mean_depth('input.bam', 'chr1', 1000000, 2000000)
print(f'Mean depth: {depth:.1f}x')

Coverage Statistics

Goal: Compute coverage breadth and depth for a genomic region from a BAM file.

Approach: Iterate pileup columns in the region, count covered positions and accumulate depth, then derive percentages and means.

Reference (pysam 0.22+):

import pysam

def coverage_stats(bam_path, chrom, start, end):
    covered = 0
    total_depth = 0

    with pysam.AlignmentFile(bam_path, 'rb') as bam:
        for pileup in bam.pileup(chrom, start, end, truncate=True):
            covered += 1
            total_depth += pileup.n

    length = end - start
    pct_covered = covered / length * 100
    mean_depth = total_depth / length if length > 0 else 0

    return {
        'length': length,
        'covered_bases': covered,
        'pct_covered': pct_covered,
        'mean_depth': mean_depth
    }

stats = coverage_stats('input.bam', 'chr1', 1000000, 2000000)
print(f'Coverage: {stats["pct_covered"]:.1f}%')
print(f'Mean depth: {stats["mean_depth"]:.1f}x')

Insert Size Distribution

Goal: Compute the insert size distribution to assess library preparation quality.

Approach: Iterate properly paired read1 records, accumulate template lengths into a Counter, then compute summary statistics.

Reference (pysam 0.22+):

import pysam
from collections import Counter

insert_sizes = Counter()

with pysam.AlignmentFile('input.bam', 'rb') as bam:
    for read in bam:
        if read.is_proper_pair and read.is_read1 and read.template_length > 0:
            insert_sizes[read.template_length] += 1

sizes = list(insert_sizes.keys())
mean_insert = sum(s * c for s, c in insert_sizes.items()) / sum(insert_sizes.values())
print(f'Mean insert size: {mean_insert:.0f}')
print(f'Min: {min(sizes)}, Max: {max(sizes)}')

Quick Reference

| Task | Command |

|------|---------|

| Quick counts | samtools flagstat input.bam |

| Per-chrom counts | samtools idxstats input.bam |

| Full stats | samtools stats input.bam |

| Coverage summary | samtools coverage input.bam |

| Per-position depth | samtools depth input.bam |

| Mean depth | samtools depth input.bam \| awk '{sum+=$3;n++}END{print sum/n}' |

QC Thresholds Are Assay-Specific

A single "mapping rate > 95%" rule rejects valid ATAC, ChIP, RNA-seq, metagenomics, and aDNA samples. The threshold question is "is this rate normal for this assay?" not "is this rate above 95%?"

| Metric | WGS PCR-free | WGS PCR | WES | Targeted panel | Deep panel (UMI) | RNA-seq | scRNA (10x) | ATAC | ChIP | Long-read | aDNA |

|--------|--------------|---------|-----|----------------|------------------|---------|-------------|------|------|-----------|------|

| Mapping rate | >99% | >98% | >95% | >95% | >95% | >90% | >70% | >50% | >60% | >95% | 1-50% |

| Duplicate rate | <5% | 5-15% | 20-50% | 20-50% | 50-90% pre-consensus | (skip) | (use UMI) | 10-30% | 5-30% | n/a | 20-60% |

| Proper pair rate | >95% | >95% | >85% | >80% | >80% | >70% | n/a | >50% | >70% | n/a | >60% |

| Mean MAPQ | bimodal at 0/60 | bimodal | bimodal | bimodal | bimodal | bimodal incl 255 (STAR) | 0/3/255 | 30-55 | 30-55 | 30-50 | 20-40 |

| Mt fraction | 0.1-2% | 0.1-2% | <1% | <0.1% | <0.1% | varies | varies | <10% target (ENCODE ATAC-seq pipeline standards, Buenrostro 2013-style libraries) | <2% | n/a | varies |

Mean MAPQ is misleading; the distribution is bimodal (0 and aligner-max). The fraction at MAPQ >= 30 is more informative:

samtools view -c -F 2308 -q 30 in.bam   # primary, mapped, MAPQ>=30
samtools view -c -F 2308 in.bam          # primary, mapped (denominator)
# For STAR/STARsolo, use -q 255 instead of -q 30 (255 is the unique-mapping sentinel)

What Flagstat Does Not Reveal

A 99% flagstat mapping rate does NOT mean the data is usable. Common false-positive scenarios:

• Adapter readthrough: short fragments (insert < 2 * read_length) sequence into adapter; aligners soft-clip the adapter portion and flag the read as MAPPED. Detect:

   samtools stats input.bam | grep "bases soft-clipped"   # >5% suggests adapter contamination

• Off-target enrichment (capture/WES): detect via picard CollectHsMetrics PCT_OFF_BAIT or PCT_SELECTED_BASES.
• Low-complexity pile-up: telomere/centromere reads mass at MAPQ-0; counted as mapped but useless. Detect via MAPQ distribution.
• Cross-sample contamination: detect via verifybamid2 or somalier (FREEMIX > 1% degrades somatic calling; > 5% breaks germline calling).
• Wrong reference build: a BAM aligned to GRCh37 viewed against GRCh38 looks fine to flagstat but produces nonsense pileups. Compare @SQ M5: from BAM header with samtools dict ref.fa -- see alignment-validation.

Insert Size Caveats

samtools stats reports the IS section only for FR-oriented properly paired reads. So:

• Mate-pair libraries (RF orientation): IS section *empty* -- proper-pair flag not set for RF
• ATAC-seq: bimodal/multimodal expected (nucleosome ladder ~50/~180/~340 bp). Unimodal suggests poor transposition.
• RNA-seq: TLEN includes intron span -- mean meaningless
• Bisulfite (PBAT): orientation reversed; samtools may not flag proper pair

Related Skills

• sam-bam-basics - View alignment files; aligner-aware MAPQ semantics
• alignment-indexing - idxstats requires index; secondary+supp counted
• alignment-validation - Insert size by library, contamination, sample-swap detection
• duplicate-handling - Library-aware duplicate rate expectations
• alignment-filtering - Filter before stats
• sequence-io/sequence-statistics - FASTA/FASTQ statistics