Miller

Miller generates small, representative mzML files from full-sized proteomics mzML datasets. Production mzML files are often hundreds of megabytes or several gigabytes — too large to bundle in repositories, share casually, or iterate on quickly. Miller solves this by extracting a configurable subset of spectra into a new, fully valid mzML file that preserves the structure and metadata of the original.

Miller works with both DDA and DIA data and is useful in a variety of scenarios:

• Smoke-testing data analysis pipelines — generate tiny mzML files to verify that a workflow runs end-to-end before committing to a full-scale run.
• CI and integration tests — ship realistic test fixtures without multi-GB raw data.
• Filtering step in a larger workflow — use Miller as a pre-processing stage, for example to trim mzML files in a cascade search or to focus on a retention-time window of interest.

Highlights

• Include or exclude scans based on scan number or retention-time range.
• Operate on specific MS levels (e.g. MS1, MS2).
• Precursor inclusion (default on) — if an MSn scan references a precursor via spectrumRef, the full precursor chain is included automatically.
• Preserves run-level sections and metadata; updates spectrumList/@count and renumbers spectrum index attributes so the output is a valid, random-access-safe indexed mzML (works with index-based readers such as MSFTBX / PDV / FragPipe).
• Recalculates TIC (MS:1000235) and BPC (MS:1000628) from retained spectra when present.
• Indexed or non-indexed mzML output, defaulting to the source unless overridden.
• Binary array compression control: source, zlib, or none.

Installation

pip (recommended)

pip install miller-mzml-filterer

Verify:

miller --help

Docker

docker pull ghcr.io/mriffle/miller-mzml-filterer:latest

Verify:

docker run --rm ghcr.io/mriffle/miller-mzml-filterer:latest --help

Quick Start

Using pip

Randomly select 50 scans:

miller --scan-count 50 input.mzML output.mzML

Randomly select 5% of scans:

miller --scan-percent 5 input.mzML output.mzML

Select 10 random MS2 scans (precursor MS1 scans are included automatically):

miller --ms-level 2 --scan-count 10 input.mzML output.mzML

Keep scans in a retention-time window:

miller --rt-range-start 35.2 --rt-range-end 35.8 input.mzML output.mzML

Select specific scans from an include file (one scan ID per line):

miller --scan-include-file scans.txt input.mzML output.mzML

Using Docker

All Docker examples below mount the current directory into the container and run as your current user/group so output files have the correct ownership:

docker run --rm \
  --user "$(id -u):$(id -g)" \
  -v "$PWD:/work" \
  -w /work \
  ghcr.io/mriffle/miller-mzml-filterer:latest \
  --scan-count 50 input.mzML output.mzML

Select 5% of scans:

docker run --rm \
  --user "$(id -u):$(id -g)" \
  -v "$PWD:/work" \
  -w /work \
  ghcr.io/mriffle/miller-mzml-filterer:latest \
  --scan-percent 5 input.mzML output.mzML

Select 10 random MS2 scans:

docker run --rm \
  --user "$(id -u):$(id -g)" \
  -v "$PWD:/work" \
  -w /work \
  ghcr.io/mriffle/miller-mzml-filterer:latest \
  --ms-level 2 --scan-count 10 input.mzML output.mzML

More Examples

Retention-time filtering

Combine an RT window with random selection:

miller --rt-range-start 35.2 --rt-range-end 35.8 --scan-count 50 input.mzML output.mzML

Pick a random contiguous 10% RT window, then select 25 scans from it:

miller --rt-window-percent 10 --scan-count 25 input.mzML output.mzML

Excluding scans

Exclude specific scans by ID (one per line in the file):

miller --scan-count 50 --scan-exclude-file exclude.txt input.mzML output.mzML

Keep all scans except the excluded ones:

miller --scan-exclude-file exclude.txt input.mzML output.mzML

Output format and compression

Force indexed output with zlib compression:

miller --indexed --compression zlib --scan-count 10 input.mzML output.mzML

Non-indexed, uncompressed:

miller --no-index --compression none --scan-count 10 input.mzML output.mzML

Precursor inclusion

By default, Miller follows spectrumRef links to include precursor scans (e.g. MS1 parents of selected MS2 scans). Disable this with:

miller --no-include-precursors --scan-count 10 input.mzML output.mzML

Determinism

Random selection is seeded (default 42). Vary the seed for different subsets of the same file:

miller --scan-count 50 --seed 1 input.mzML output_seed1.mzML
miller --scan-count 50 --seed 2 input.mzML output_seed2.mzML

---

CLI Reference

miller [OPTIONS] INPUT OUTPUT

Positional arguments

• INPUT — path to the source mzML file (indexed or non-indexed).
• OUTPUT — path for the output mzML file.

Selection mode (mutually exclusive)

• --scan-count INTEGER — randomly select N scans from the eligible pool. Fails if N exceeds pool size.
• --scan-percent FLOAT — randomly select a percentage (> 0, ≤ 100) of eligible scans.
• --scan-include-file PATH — file with one scan ID per line. Accepts bare numbers (1001) or prefixed IDs (scan=1001).
• If none of the above are given and --scan-exclude-file is set, all scans minus exclusions are kept.

Filtering

• --rt-range-start FLOAT / --rt-range-end FLOAT — inclusive RT bounds applied before selection. Either or both may be supplied.
• --rt-window-percent FLOAT — random contiguous RT window (percentage of eligible RT span), applied after fixed RT bounds.
• --scan-exclude-file PATH — one scan ID per line to exclude from selection and final output.
• --ms-level TEXT — comma-separated MS levels (e.g. 1, 2, 1,2). Valid only with --scan-count or --scan-percent.

Precursor inclusion

• --include-precursors / --no-include-precursors (default: include) — walk spectrumRef chains to include ancestor scans.

Output format

• --indexed / --no-index — force indexed or non-indexed output. Default follows the source file.
• --compression [source|zlib|none] (default: source) — binary array compression mode.

Other

• --seed INTEGER (default: 42) — random seed for --scan-count, --scan-percent, and --rt-window-percent.
• --help / -h — show usage and exit.
• --version / -v — show version and exit.

Exit codes

Code	Meaning
1	Invalid or unreadable input file
2	CLI usage / argument error
3	One or more explicit scan IDs not found
4	Selection produced zero eligible scans
5	Output path / write error

---

Development

Local setup

python3 -m venv .venv
. .venv/bin/activate
pip install -e ".[dev]"

Running tests

.venv/bin/pytest --cov=miller --cov-report=term-missing tests/
.venv/bin/ruff check src/ tests/
.venv/bin/mypy src/

Smoke tests use test_data/test_data.mzML and run automatically with the full suite. To run only smoke tests:

.venv/bin/pytest tests/test_smoke_real_data.py

Building the Docker image locally

docker build -t miller .
docker run --rm miller --help

Running tests inside Docker

docker run --rm --entrypoint pytest ghcr.io/mriffle/miller-mzml-filterer:latest \
  --cov=miller --cov-report=term-missing tests/