lesSDRF

This github repository contains the code used to create the lesSDRF application which is available via https://lessdrf.streamlit.app/ A manual on the tool in pdf format is available here: https://github.com/compomics/lesSDRF/blob/main/lesSDRF_manual.pdf

lesSDRF v0.2.0 follows the SDRF-Proteomics specification v1.1.0. The column definitions, allowed values and validation come from the templates bundled with sdrf-pipelines (the official validator), so upgrading sdrf-pipelines in requirements.txt keeps the app in line with the specification.

Running the app

Python 3.10 or newer is required (sdrf-pipelines needs it); the app is tested with Python 3.11.

pip install -r requirements.txt
streamlit run Home.py

This opens the Home screen of the app. The following steps can be found in the pages folder and are numbered accordingly. On Streamlit Community Cloud, select Python 3.11 (or newer) in the app settings.

Alternatively, during development, you can use docker compose to build and run the app using the watch feature locally:

docker compose watch

This will automatically sync file changes from your local machine to the docker container. Streamlit will automatically pick up file changes and will offer to reload the application.

Open your browser and go to http://localhost:8501 to see the app running.

Code structure

File	Purpose
Home.py, pages/	The Streamlit pages (one per step)
spec.py	Everything that depends on the SDRF specification: templates, column rules, value formats, column order, export and validation
app_utils.py	Shared Streamlit building blocks: data loading, ontology search, editors and the sidebar
ParsingModule.py, *.ipynb	Offline helpers and notebooks that turn ontology files into the data files in data/
textmining.py	Finds SDRF metadata in pasted methods text (step 1, From methods text)
scripts/build_accession_map.py	Builds data/term_accessions.json.gz (label → PSI-MS/PRIDE accession)

Metadata from methods text

Step 1 can read a pasted methods section and suggest SDRF values, without an LLM or any external service: dictionary matching against the bundled ontologies (diseases, tissues, cell lines, instruments, ...), rules for common phrasings (enzymes, reagents, modifications and whether they are fixed or variable, TMT/iTRAQ/SILAC channels, DDA/DIA, HCD, collision energy, mass tolerances, age, sex) and NCBITaxon look-ups for Latin species names. Every suggestion is highlighted in the text, has a confidence score and must be accepted before it is written to the SDRF.

The confidence scores are calibrated on the 107 manuscripts with reference SDRFs in the agentic-metadata benchmark: suggestions scored 0.6–0.7 were correct about 65% of the time, 0.7–0.8 about 75% and above 0.8 just over 80%. Study-wide parameters are found best (cleavage agent, organism, dissociation method, modifications, instrument); sample-level columns such as disease and organism part are often absent from methods text or ambiguous, so their suggestions get low scores.

Memory and speed

The hosted app has limited memory. To keep it stable with many simultaneous users:

• Ontology data is loaded lazily, one file at a time, and shared between all sessions (st.cache_resource); nothing large is stored per session.
• Large term lists (e.g. 137k diseases, 1.7M NCBITaxon names) are searched on the server; only matches are sent to the browser. Ontology trees are only shown on request and are pruned to ~1.5 MB.
• The SDRF is validated in-process with sdrf-pipelines, only when you click Validate SDRF.

Ontology data

The parsed ontologies used in the app are in the data folder. Every topic contains three types of data:

• all_elements: a list of every term in the ontology subset. Is used to check for ontology compatibility of local metadata.
• dict: a nested dictionary that follows the ontology tree structure
• nodes: a node like version of the nested dictionary according to the format required for the tree-select module: https://github.com/Schluca/streamlit_tree_select

data/term_accessions.json.gz maps PSI-MS and PRIDE term labels to accessions, so that comment columns are written as NT=<label>;AC=<accession> as the specification recommends. Regenerate it after updating the ontology files:

python scripts/build_accession_map.py ontology/psi-ms20251015.obo ontology/pride_cv20251015.obo

Ontology versions

The term lists in data/ were last refreshed on 2025-10-15 (EFO-based lists such as disease, cell line and developmental stage); see the git history of data/ for earlier updates. The accession map is built from PRIDE CV (release 2025-02-17) and PSI-MS (4.1.205).