ProSE in the browser

The OpenMS search engine ProSE, compiled to WebAssembly, with Percolator rescoring and, optionally, PeptDeep predictions of fragment intensities and retention times (ONNX Runtime). The search runs entirely in your browser: your spectra and database stay on your machine.

Open the application

Try it with example data

1. DatabaseSuitability_in_spec.mzML (6.5 MB) with DatabaseSuitability_database.fasta: set fragment mass tolerance to 10, then Search. About 900 PSMs at 1 % FDR; most of the time goes to Percolator.
2. Smaller: SimpleSearchEngine_1.mzML with SimpleSearchEngine_1.fasta (3 PSMs; Percolator is skipped, it needs at least 100).
3. Your own mzML and FASTA files.

Click a PSM for the annotated spectrum, the fragment coverage, the mass errors and all PSM features.

Tick with PeptDeep features next to Rescore with Percolator to add PeptDeep (AlphaPeptDeep) predictions: the agreement of the predicted fragment intensities and retention time with each PSM becomes five more Percolator features (ms2_cosine, ms2_spectral_angle, ms2_pearson, ms2_frac_pred_found, rt_abs_error). The models (17 MB) load on first use. The example data is too small and synthetic to gain from it; real label-free HCD data does (+5.7 % PSMs at 1 % FDR in the OpenMS benchmark).

Desktop application

The same application as a small program that works offline, with all threads, in a window of your Chrome, Chromium or Edge, PeptDeep option and models included (packaged with Neutralinojs, about 30 MB):

• Windows (x64)
• macOS (Intel and Apple silicon)
• Linux (x64) · Linux (arm64)

Unzip, keep the program and resources.neu together and start the program; closing the window ends it. The programs are not code-signed, so Windows SmartScreen and macOS Gatekeeper warn at first: README.txt in the zip says how to start them anyway. Tested on Linux; the Windows and macOS packages are untested.

Notes

• Tested in Chrome and Edge; Firefox and Safari are untested.
• On the first visit the page reloads once: a service worker makes it cross-origin isolated, which threads need, and caches the 30 MB module. Later visits start faster and search faster, because the browser keeps the compiled module. A hard reload (Shift+Reload) bypasses the service worker.
• The module can use 4 GB of memory. For large searches (e.g. semi-specific against a large database), set database:chunk_size under All parameters. PeptDeep needs about 0.6 GB more.
• Without a server: prose-single-file.html (9.6 MB) is the same application in one file that opens from disk, on one thread (a page opened from a file cannot be cross-origin isolated), without PeptDeep.
• The general OpenMS demo: peptide masses, digestion, an mzML viewer.

A preview built from OpenMS pull request #10575 (commit 80a3b501c4, threaded build with ONNX Runtime), not an OpenMS release. Source of this page · JavaScript API · License.