Gensim pipeline execution

gensim-pipeline-execution

Summary

Execute a three-stage gensim LDA pipeline to infer topics from LC-MS/MS metabolomics data: generating a corpus from MS2 spectra, running latent Dirichlet allocation, and loading results into PostgreSQL. This skill orchestrates the complete workflow from raw mass spectrometry input to database-backed topic visualization.

When to use

Apply this skill when you have an MS2 spectral file (MGF format) from LC-MS/MS metabolomics analysis and need to discover latent topics across fragmentation patterns for unsupervised characterization. Use it as the prerequisite step before topic visualization in the ms2lda web interface.

When NOT to use

• Input is already a pre-computed LDA model or topic table — use the insert endpoint directly instead.
• MS2 file is in a format other than MGF — first convert to MGF or use appropriate preprocessing.
• PostgreSQL or Django server is not running — set up Docker containers or local services first.

Inputs

• MS2 spectral file (MGF format)
• PostgreSQL server connection (running and accessible)
• Django server (up and running, as noted in README)

Outputs

• Corpus JSON file (intermediate: feature representation from MS2 data)
• LDA result JSON file (gensim model: topic distributions and model parameters)
• PostgreSQL database records (experiment metadata, topic assignments, LDA artifacts)

How to apply

Execute three sequential commands via run_gensim.py: (1) corpus generation from an MGF file with ./run_gensim.py corpus -f mgf <input.mgf> <output.corpus.json>, which extracts features/corpus representation; (2) gensim LDA inference with ./run_gensim.py gensim <corpus.json> <ldaresult.json>, which infers topic distributions; (3) database insertion with ./run_gensim.py insert <ldaresult.json> <username> <experiment_name>, which loads the JSON result into PostgreSQL for web access. Verify success by confirming no errors in each stage and that the experiment is accessible in the database.

Related tools

• gensim (Performs Latent Dirichlet Allocation topic inference on the corpus representation of MS2 spectra)
• PostgreSQL (Stores LDA results, experiment metadata, and topic assignments for persistent access and web visualization)
• Django (Web application framework hosting the ms2lda server that provides access to inserted LDA experiments) — https://github.com/glasgowcompbio/ms2ldaviz
• Python (Language in which run_gensim.py and the pipeline orchestration are implemented)
• Docker (Containerization for PostgreSQL and other services required by the ms2lda application)

Examples

./run_gensim.py corpus -f mgf myexp.mgf myexp.corpus.json && ./run_gensim.py gensim myexp.corpus.json myexp.ldaresult.json && ./run_gensim.py insert myexp.ldaresult.json stefanv myexp

Evaluation signals

• No error messages or exceptions from any of the three run_gensim.py commands (corpus, gensim, insert).
• Corpus JSON file is valid JSON and contains topic term distributions matching the input MGF feature count.
• LDA result JSON file is parseable and includes model parameters (alpha, beta, topic distributions).
• PostgreSQL query confirms that the experiment record exists with the provided username and experiment name.
• Web interface (http://localhost:8000) returns the experiment when queried or browsed post-insertion.

Limitations

• Requires a running Django server and PostgreSQL database — offline or disconnected workflows will fail at the insert stage.
• The pipeline is synchronous and blocking; large MGF files may require significant computation time for LDA inference.
• Input must be MGF format; other MS2 formats (e.g., mzML) require external conversion before corpus generation.

Evidence

• [intro] topics inferred from Latent Dirichlet Allocation can be used to assist in the unsupervised characterisation of fragmented (LC-MS-MS) metabolomics data: "topics inferred from Latent Dirichlet Allocation can be used to assist in the unsupervised characterisation of fragmented (LC-MS-MS) metabolomics data"
• [readme] Performs 3 steps: 1. Generate corpus/features from MS2 file 2. Run lda using gensim 3. Insert lda result into db: "Performs 3 steps: 1. Generate corpus/features from MS2 file 2. Run lda using gensim 3. Insert lda result into db"
• [readme] ./run_gensim.py corpus -f mgf myexp.mgf myexp.corpus.json ./run_gensim.py gensim myexp.corpus.json myexp.ldaresult.json ./run_gensim.py insert myexp.ldaresult.json stefanv myexp: "./run_gensim.py corpus -f mgf myexp.mgf myexp.corpus.json ./run_gensim.py gensim myexp.corpus.json myexp.ldaresult.json ./run_gensim.py insert myexp.ldaresult.json stefanv myexp"
• [readme] Requires server to be up and running.: "Requires server to be up and running."
• [other] The insert step consumes the LDA result JSON file (myexp.ldaresult.json) and loads it into the database using the command ./run_gensim.py insert with parameters for the result file, a username (stefanv), and an experiment name (myexp).: "The insert step consumes the LDA result JSON file (myexp.ldaresult.json) and loads it into the database using the command ./run_gensim.py insert with parameters for the result file, a username"