May 27, 2026
Fine-tuning tagged spectral libraries in DIA-NN
This protocol is a draft, published without a DOI.
- Harrison Specht1
- 1Parallel Squared Technology Institute
Protocol Citation: Harrison Specht 2026. Fine-tuning tagged spectral libraries in DIA-NN. protocols.io https://dx.doi.org/
License: This is an open access protocol distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Protocol status: Working
We use this protocol and it's working
Created: May 26, 2026
Last Modified: May 27, 2026
Protocol Integer ID: 317953
Keywords: DIA-NN, fine tuning, fine-tuning, spectral library, PSMtag, proteomics, proteomics search software dia, release of proteomics search software dia, specific spectral library predictions for psmtag, proteomics search software, useful tool for proteomic, spectral libraries in dia, tuning spectral library prediction, peptide modification, tagged spectral library, specific spectral library prediction, proteomic, modified peptide, spectral library prediction, empirical spectral library, psmtag data, mass tag mtraq, accurate proteome, modification by mass tag, using dia, mass tag, alphapeptdeep by the mann lab, psmtag, earlier versions of dia, models for several modification
Funders Acknowledgements:
Eric and Wendy Schmidt
Griffin Catalyst
Disclaimer
Prof. Demichev has noted that building fragmentation models is not without the pit-falls common to any model-building process, such as overtraining and not properly controlling false discovery rate.
Abstract
TL;DR: We have made some models so now you can predict libraries for PSMtag'd peptides. Some are here but you can make your own following the tutorial below. Credit to Prof. Vadim Demichev for creating this highly useful tool for proteomics.
After the release of proteomics search software DIA-NN 2.3.0 by Prof. Vadim Demichev, researchers can now fine-tune spectral library predictions for a range of peptide modifications, including modification by mass tags like PSMtag. Models for several modifications, including the mass tag mTRAQ, have been built into earlier versions of DIA-NN. Another proteomics search software, AlphaPeptDeep by the Mann lab, has allowed for fine-tuning spectral library predictions of modified peptides, as well. Here, we will offer a tutorial for how to build instrument-specific spectral library predictions for PSMtag modified peptides using DIA-NN 2.3.0. This will allow users to do more accurate proteome-wide searches of their PSMtag data and potentially by-pass the need to first generate empirical spectral libraries from bulk or fractionated samples.
Before start
Note 1: When DIA-NN opens a new window, a set of default options are loaded. In this tutorial we err on the side of changing as few of these default options as needed, since DIA-NN will turn many of them off automatically if they are not necessary for a given step. For example, we leave MBR checked, but it will automatically turn off for steps involving none or one raw file.
Note 2: If your PSMtag data is multiplexed, add the appropriate “--channels …” command with the appropriate channel masses to Steps 3 and 6.
Additional options for Steps 3 and 6 if data is 5-plexDIA:
-channels tag,d0,nK,0.0; tag,d4,nK,4.01096;4.01096; tag,d8,nK,8.026839;8.026839; tag,d12,nK,12.033938;12.033938; tag,d16,nK,16.038574;16.038574
Tutorial steps
Acquire PSMtag'd proteomic data by data-independent acquisition (DIA) on your instrument of choice. We offer the peptide-labeling protocol and suggest LC-MS methods for the Thermofisher Astral and Bruker timsTOF platforms.
Generate a predicted spectral library using the default (label-free) models and your fasta of choice.
Your log file should look like this at Step 2:
diann.exe --lib "" --threads 16 --verbose 3 --out "L:\HS\tutorial\report.parquet" --qvalue 0.01 --matrices --out-lib "L:\HS\tutorial\report-lib.parquet" --gen-spec-lib --predictor --fasta "L:\fasta\proteome.fasta" --fasta-search --met-excision --min-pep-len 7 --max-pep-len 30 --min-pr-mz 300 --max-pr-mz 1800 --min-pr-charge 1 --max-pr-charge 4 --min-fr-mz 200 --max-fr-mz 1800 --cut K*,R* --missed-cleavages 1 --unimod4 --reanalyse --rt-profiling
Add PSMtag modification to the library.
Your log file should look like this at Step 3:
diann.exe --lib "L:\HS\tutorial\report-lib.predicted.speclib" --threads 16 --verbose 3 --out "L:\HS\tutorial\report.parquet" --qvalue 0.01 --matrices --out-lib "L:\HS\tutorial\report-lib.parquet" --gen-spec-lib --unimod4 --reanalyse --rt-profiling --fixed-mod tag, 308.1160923903, nK --lib-fixed-mod tag
Search your data with relaxed-constraints.
Your log file should look like this at Step 4:
diann.exe -f "L:\HS\20250520_PSMtag_share\bulkdata-astral\2025-04-28_PSMtag_doanhy_20ng_5Th-DIA.raw " --lib "L:\HS\tutorial\report-lib.parquet" --threads 16 --verbose 3 --out "L:\HS\tutorial\pretuned_search\report.parquet" --qvalue 0.01 --matrices --min-corr 2.0 --time-corr-only --extracted-ms1 --min-cal 500 --min-class 1000 --pre-filter --no-rt-window --out-lib "L:\HS\tutorial\pretuned_search\report-lib.parquet" --gen-spec-lib --unimod4 --proteoforms --reanalyse --fixed-mod tag, 308.1160923903, nK
Train fragment, retention time and ion mobility (optional) models.
Your log file should look like this at Step 5:
diann.exe --lib "" --threads 16 --verbose 3 --out "C:\DIA-NN2.3.0\report.parquet" --qvalue 0.01 --matrices --out-lib "C:\DIA-NN2.3.0\report-lib.parquet" --gen-spec-lib --unimod4 --reanalyse --rt-profiling --tune-lib "L:\HS\tutorial\pretuned_search\report-lib.parquet --tune-rt --tune-fr --fixed-mod tag, 308.1160923903, nK --original-mods
Tune the library generated in Step 3 with the models generated in Step 5.
Note: If the models don’t seem to load, try renaming the files to remove the “_” and extra “.”
Your log file should look like this at Step 6:
diann.exe --lib "L:\HS\tutorial\report-lib.parquet" --threads 16 --verbose 1 --out "L:\HS\tutorial\tuned_lib\report.parquet" --qvalue 0.01 --matrices --out-lib "L:\HS\tutorial\tuned_lib\report-lib.parquet" --gen-spec-lib --predictor --fasta "L:\fasta\proteome.fasta" --met-excision --min-pep-len 5 --max-pep-len 30 --min-pr-mz 0 --max-pr-mz 1800 --min-pr-charge 1 --max-pr-charge 5 --min-fr-mz 0 --max-fr-mz 1800 --cut K*,R* --missed-cleavages 1 --unimod4 --reanalyse --rt-profiling --tokens L:\HS\tutorial\pretuned_search\dict.txt --rt-model L:\HS\tutorial\pretuned_search\rt.pt --fr-model L:\HS\20251015_tuning_dia_s1\fr.pt --fixed-mod tag, 308.1160923903, nK --original-mods
Apply the fine-tuned library to other PSMtag data.
Your log file should look like this at Step 7:
diann.exe -f "L:\HS\20250520_PSMtag_share\bulkdata-astral\2025-04-28_PSMtag_doanhy_20ng_5Th-DIA_rep2.raw " --lib "L:\HS\tutorial\tuned_lib\report-lib.predicted.speclib" --threads 16 --verbose 3 --out "L:\HS\tutorial\posttuned_search\report.parquet" --qvalue 0.01 --matrices --out-lib "L:\HS\tutorial\posttuned_search\report-lib.parquet" --gen-spec-lib --fasta "L:\fasta\proteome.fasta" --met-excision --min-pep-len 5 --max-pep-len 30 --min-pr-mz 0 --max-pr-mz 1800 --min-pr-charge 1 --max-pr-charge 5 --min-fr-mz 0 --max-fr-mz 1800 --cut K*,R* --missed-cleavages 1 --unimod4 --reanalyse --rt-profiling --fixed-mod tag, 308.1160923903, nK --original-mods
P.S. If you are savvy with DIA-NN's pipeline building function, you can automate all above Steps by changing the paths and filenames in the pipeline included here.
Acknowledgements
Prof. Vadim Demichev, for creating the DIA-NN software.