visDIA: MuMDIA results viewer

mumdia-viewer is an interactive, read-only viewer for the outputs of MuMDIA, a DIA proteomics search engine. It opens a run or an experiment directory and shows the identifications, the evidence behind each one (XICs, spectra, decoy competition), the calibrations, the quantification, and the differences between two result sets.

Status: all views of the specification are built: the run overview, the identification browser and the precursor detail (P0); calibration, run QC, quant QC and the protein view (P1); experiment views, compare, the spectrum browser, export and validation notes (P2). The viewer reads MuMDIA v0.5.0 outputs, and the schema versions of earlier releases. It never writes to a run directory.

Install

pip install .                     # or the wheel: pip install mumdia_viewer-*.whl

Install (development)

python -m venv .venv
.venv/Scripts/activate            # Windows; use `source .venv/bin/activate` elsewhere
pip install -e ".[dev]"

Python 3.11 or newer.

Run the viewer

mumdia-viewer <run-or-experiment-dir> [--fasta <proteins.fasta>] [--compare <other-dir>] [--port N]

The viewer serves at http://127.0.0.1:<port>/<token>/ and opens a browser. The random token in the address keeps other users of a shared machine out.

• FASTA: --fasta gives the protein sequences for the coverage views. A run searched directly from a FASTA records it, and the viewer then finds it without the option.
• Remote server: start it there with --no-browser, then forward the port: ssh -L <port>:127.0.0.1:<port> <user>@<server>. Open the printed address on your machine.
• Inputs that moved: --remap OLD=NEW says where inputs recorded under OLD are now.

The user guide describes every page, the options, where the viewer keeps its files, and how to solve common problems.

Every number is MuMDIA's own column. A number the viewer derives (a percentile, a coverage, a CV, a viewer-side spectrum match) says so where it is shown.

The data layer

mumdia_viewer.data is a plain Python API that returns pandas DataFrames, numpy arrays and dataclasses, for use in a notebook or another application:

from mumdia_viewer.runtime import configure_environment
configure_environment()                      # before numpy/pyarrow: bounded threads and memory

from mumdia_viewer.data import open_results
from mumdia_viewer.data import counts, tables
from mumdia_viewer.data.detail import precursor_detail, mirror

rs = open_results("path/to/run-or-experiment")
for c in counts.unit_counts(rs, 0.01):
    print(c.label)                           # e.g. "81,310 peptides (unique base_peptide_id, peptide_q_value <= 0.01)"

page = tables.identification_table(rs, tables.TableQuery(unit="precursor", limit=20))
detail = precursor_detail(rs, rs.runs[0], int(page.rows.candidate_id.iloc[0]))
spectrum = mirror(rs, detail)                # apex MS2 scan against the predicted fragments

See docs/data-layer.md for how artifacts are found, versioned, cached and counted.

Tests

pytest                                        # hermetic tests on tests/fixtures
MUMDIA_VIEWER_REAL_SINGLE=<run dir> MUMDIA_VIEWER_REAL_EXPERIMENT=<experiment dir> pytest -m real_data
python benchmarks/m1_performance.py <run dir> # open, precursor detail and memory, cold and warm
python benchmarks/m2_ui_performance.py <run dir> # server time of the pages, cold and warm

The UI tests build the pages on the fixtures without a browser. Some also drive headless Chromium through Playwright when it is installed (pip install -e ".[dev]", then playwright install chromium).

Licence: Apache-2.0.