DelPi is an open-source, easy-to-use peptide identification tool for mass spectrometry-based proteomics. It provides a Windows GUI application and a command-line interface, helping users run DIA or DDA peptide searches without building complex workflows by hand.
DelPi applies a pre-trained Transformer encoder to score candidate peptides from raw MS1/MS2 evidence using an acquisition-agnostic representation, enabling a unified workflow across both DIA and DDA data.
Run peptide identification through an easy-to-use Windows GUI — no manual YAML editing required.
- Deep representation learning: Scores candidate peptides using a pre-trained Transformer encoder, without relying on handcrafted features.
- DIA and DDA support: Use one workflow across common LC-MS/MS acquisition modes.
- Library-free search: Generates in silico spectral libraries internally, supporting common PTMs registered in the UniMod database.
- GPU-accelerated inference: Designed for practical performance on consumer-grade GPUs via PyTorch/CUDA.
- Experiment-adaptive workflow: Employs a two-stage search with experiment-level transfer learning to adapt to instrument and chromatographic conditions.
- Easy-to-use Windows GUI: Configure and run DelPi searches without manually editing YAML files.
Memory: ≥ 32 GB RAM
Compute:
- NVIDIA GPU with CUDA support required
- Supported OS: Linux, Windows
- macOS (including Apple Silicon/MPS) and CPU-only execution are not supported
Memory Considerations:
- DelPi processes input files one run at a time
- Peak memory usage depends on the size of an individual raw/mzML file being processed
- Recommended available memory: (single run file size + ~16 GB) to accommodate intermediate data structures, model execution, and OS overhead
Runtime:
- For a 25 min DIA gradient (human sample, Astral Orbitrap), DelPi completes peptide identification in approximately 15 minutes on a single NVIDIA RTX 4090 GPU
- Formats: mzML and Thermo RAW (other vendor formats can be converted to mzML using ProteoWizard MSConvert)
- Acquisition modes: DIA and DDA
- Ion mobility (IM) data (e.g., FAIMS, PASEF) is not currently supported, but will be supported soon
DelPi is available in two flavors:
- Windows GUI application — recommended for Windows users who prefer a graphical interface.
- Command-line tool — cross-platform (Linux/Windows) installation from source.
For most Windows users, this is the easiest way to start using DelPi. Download the latest Windows installer (.exe) from the Releases page and run it. The installer bundles all required dependencies; no additional setup is needed.
Windows users: Please use PowerShell (not Command Prompt/cmd) for all installation steps. The install scripts (
.ps1) require PowerShell to run.
git clone https://github.com/bertis-informatics/delpi.git
cd delpiCreate a virtual environment using venv (Option A) or conda (Option B). Package installation in later steps uses pip or uv.
# Create a virtual environment with Python 3.12
uv venv delpi_env --python 3.12
# If not using uv: python -m venv delpi_env
# Activate the virtual environment
# Windows:
delpi_env\Scripts\activate
# macOS/Linux:
source delpi_env/bin/activate# Create a conda environment with Python 3.12
conda create -n delpi_env python=3.12 -y
# Activate the environment
conda activate delpi_envVisit the PyTorch official website to obtain the appropriate installation command for your system.
Example for CUDA 12.8:
uv pip install torch --index-url https://download.pytorch.org/whl/cu128
# If using pip: pip install torch --index-url https://download.pytorch.org/whl/cu128pymsio is bundled in the pymsio/ directory. The install script downloads the Thermo RawFileReader DLLs and installs pymsio in one step. For additional details, see the pymsio README.
Windows PowerShell:
.\pymsio\install.ps1Linux:
chmod +x pymsio/install.sh
./pymsio/install.shuv pip install .
# If using pip: pip install .delpi --help
python -c "import delpi; print('DelPi installed successfully!')"Verify your DelPi installation using publicly available DIA data from the Skyline tutorial.
wget https://skyline.ms/tutorials/DIA-QE.zip
unzip DIA-QE.zip- Open the DelPi application.
- Select the downloaded raw or mzML input files.
- Select a FASTA protein database.
- Choose an output folder for search results.
- Choose an output spectral library folder.
- Review the search settings.
- Click Run.
- Check the result files in the output folder.
-
Configure search parameters:
Copy the example configuration file:
cp data/example_param.yaml my_config.yaml
Edit
my_config.yamlto specify paths forinput_files,fasta_file,output_directory, anddatabase_directory. -
Run the search:
delpi my_config.yaml
-
Verify output:
DelPi generates the following files in your specified
output_directory:delpi.log: Detailed execution logpmsm_results.<tsv|parquet>: Peptide-spectrum matches with q-values (format depends on configuration; example)protein_group_maxlfq_results.tsv: MaxLFQ protein quantification (example)
Compare your results with the provided examples to verify correct installation.
Ensure your data files are in a supported format. If needed, convert to mzML using ProteoWizard MSConvert.
Create a YAML configuration file based on the example template.
Required fields:
| Field | Description |
|---|---|
| acquisition_method | Acquisition mode (DIA or DDA) |
| input_files | Paths to LC–MS/MS data files. Accepts a single string or a list, where each entry is either an explicit file path or a glob pattern (*, ?, [], and recursive ** are supported), e.g. /data/*.mzML or /data/**/*.mzML. |
| fasta_file | Protein database in FASTA format |
| output_directory | Directory where search results will be written |
| database_directory | Directory for storing internally generated in silico spectral libraries (if libraries generated using the same FASTA file and search options already exist, they will be reused) |
Optional fields:
Digestion and modification parameters can be adjusted for your experimental setup. Modifications can be specified either by their PSI-MS controlled vocabulary names (e.g., Oxidation, Carbamidomethyl) or by their UniMod accession numbers in the UniMod:XX format (e.g., UniMod:35, UniMod:4).
Execute DelPi with your configuration file:
delpi /path/to/your/config.yamlDelPi generates the following output files:
Click to expand output fields
| Field name | Description |
|---|---|
| frame_num | Scan number corresponding to the center of the Peptide–Multi-Spectra Match (PmSM) |
| run_name | Name of the LC–MS run |
| modified_sequence | Peptide sequence including post-translational modifications |
| precursor_charge | Charge state of the precursor ion |
| sequence_length | Length of the peptide sequence |
| is_decoy | Indicator specifying whether the match originates from a decoy sequence |
| predicted_rt | Predicted retention time of the peptide |
| observed_rt | Observed retention time of the peptide |
| score | Raw PmSM score assigned by the DelPi scoring model |
| global_precursor_q_value | Global precursor-level q-value across all runs |
| global_peptide_q_value | Global peptide-level q-value across all runs |
| global_protein_group_q_value | Global protein group-level q-value across all runs |
| protein_group | Protein group inferred according to the parsimony principle (FASTA IDs separated by semicolons) |
| fasta_id | FASTA IDs associated with the peptide, separated by semicolons |
| precursor_q_value | Run-specific precursor-level q-value |
| peptide_q_value | Run-specific peptide-level q-value |
| protein_group_q_value | Run-specific protein group-level q-value |
| ms1_quantity | Integrated area under the precursor ion chromatogram in MS1 spectra |
| ms2_quantity | (DIA only, optional) Precursor abundance quantified from fragment-level signals, before run/RT-dependent normalization |
| ms2_quantity_normalized | (DIA only, optional) ms2_quantity after run/RT-dependent normalization across runs; used as the input to MaxLFQ protein-group quantification |
Protein-level quantification results report (DIA only, optional) protein_group_maxlfq_results.<tsv|parquet>
Click to expand output fields
| Field name | Description |
|---|---|
| run_name | Name of the LC–MS run |
| protein_group | Protein group inferred according to the parsimony principle (FASTA IDs separated by semicolons) |
| maxlfq_abundance | Protein abundance calculated using the MaxLFQ algorithm (Cox et al., 2014) from normalized precursor quantities (ms2_quantity_normalized) |
If you use DelPi in your research, please cite:
Park, J., Kim, K., Kang, U.-B., & Kim, S. DelPi Learns Generalizable Peptide–Signal Correspondence for Mass Spectrometry-Based Proteomics. bioRxiv (2026). https://doi.org/10.64898/2026.01.06.697814
DelPi is freely available under the MIT License.
For questions, bug reports, or feature requests, please contact Jungkap Park, Ph.D. at jungkap.park@bertis.com