Open-Source Peptide-Centric Search Tool Designed for Skyline Integration
Osprey is an open-source tool for peptide detection and quantification in data-independent acquisition (DIA) mass spectrometry data. It uses fragment XIC co-elution analysis to detect peptides in DIA data, with machine learning scoring and rigorous FDR control.
- Peptide-centric analysis: Directly scores peptide candidates against observed spectra
- Fragment co-elution: Extracts fragment XICs, computes pairwise correlations, and scores using spectral matching at the apex
- High-resolution support: ppm-based fragment matching for high-resolution instruments (Astral, Orbitrap)
- RT calibration: LOESS-based retention time calibration with stratified sampling
- Decoy generation: Enzyme-aware sequence reversal with fragment m/z recalculation
- Skyline integration: Outputs BiblioSpec (.blib) format for seamless quantification in Skyline
- FDR control: Built-in Percolator-style semi-supervised SVM with two-level FDR (run + experiment) at precursor and peptide levels
- Protein FDR: Native protein parsimony with true picked-protein FDR (Savitski 2015), shared peptide handling (All/Razor/Unique), and two-pass architecture (first-pass protein q-values gate compaction and reconciliation consensus; second-pass q-values are authoritative)
- Cross-run reconciliation: Consensus RT alignment across replicates with peak boundary imputation for missing detections; tolerance derived from within-peptide RT reproducibility (typically 3-5x tighter than cross-peptide calibration MAD)
- Peptide trace diagnostics:
OSPREY_TRACE_PEPTIDE=<sequence>emits a detailed per-peptide log at every pipeline stage (CWT candidates, consensus, reconciliation, gap-fill, FDR) for investigating integration quality on specific peptides. See docs/17-peptide-trace.md. - Flexible input: Supports DIA-NN TSV, EncyclopeDIA elib, and BiblioSpec blib libraries
- Scalable: Disk-backed memory architecture with automatic cache invalidation; tested on 240-file experiments
- Configurable: YAML configuration files with CLI argument overrides for reproducible analyses
- Rust 1.75 or later
- OpenBLAS development libraries
- CMake
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source "$HOME/.cargo/env"Follow the on-screen prompts (the defaults are fine). After installation, cargo and rustc will be available in your terminal.
sudo apt-get update
sudo apt-get install build-essential libopenblas-dev cmake pkg-configgit clone https://github.com/maccoss/osprey.git
cd osprey
cargo build --release
cargo install --path crates/ospreyDownload and run rustup-init.exe. The default installation (MSVC toolchain) is recommended.
Rust on Windows requires the MSVC C/C++ build tools. If you don't have Visual Studio installed, download Visual Studio Build Tools and install the "Desktop development with C++" workload.
Download and install CMake. During installation, select "Add CMake to the system PATH".
Alternatively, install via winget:
winget install Kitware.CMakeOsprey requires OpenBLAS for linear algebra operations. Download pre-built binaries from the OpenBLAS releases page (choose the latest OpenBLAS-*-x64.zip).
Extract the archive and set the environment variable so the build system can find it:
# Example: extracted to C:\OpenBLAS
[System.Environment]::SetEnvironmentVariable("OPENBLAS_PATH", "C:\OpenBLAS", "User")Alternatively, install OpenBLAS via vcpkg:
git clone https://github.com/microsoft/vcpkg.git
cd vcpkg
.\bootstrap-vcpkg.bat
.\vcpkg install openblas:x64-windows
# Set the environment variable to the installed path
[System.Environment]::SetEnvironmentVariable("OPENBLAS_PATH", "$PWD\installed\x64-windows", "User")After setting environment variables, restart your terminal for changes to take effect.
git clone https://github.com/maccoss/osprey.git
cd osprey
cargo build --release
cargo install --path crates/ospreyThe built binary will be at target\release\osprey.exe, and cargo install places it in %USERPROFILE%\.cargo\bin\ (which rustup adds to PATH automatically).
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source "$HOME/.cargo/env"brew install openblas cmake
export OPENBLAS_PATH="$(brew --prefix openblas)"Add the OPENBLAS_PATH export to your ~/.zshrc (or ~/.bash_profile) so it persists across sessions.
git clone https://github.com/maccoss/osprey.git
cd osprey
cargo build --release
cargo install --path crates/ospreyOsprey includes a built-in Percolator-style semi-supervised SVM — no Python required. Mokapot is an optional alternative FDR engine that can be enabled with --fdr-engine mokapot.
# Requires Python 3.8+ and pip
# Note: mokapot 0.10.0 requires pandas 2.x and numpy <2.0
pip install mokapot 'pandas>=2.0,<3.0' 'numpy<2.0'If the mokapot command is not found after installation, add the scripts directory to your PATH:
- Linux: Add
~/.local/binto PATH in~/.bashrc - Windows:
pip install --userplaces scripts in%APPDATA%\Python\PythonXX\Scripts— add this to your system PATH, or usepip install(without--user) in an activated virtual environment - macOS: Add
~/Library/Python/X.Y/binto PATH
# Analyze DIA data with a spectral library (default: 10 ppm fragment tolerance)
osprey -i sample.mzML -l library.tsv -o results.blib
# Multiple input files
osprey -i *.mzML -l library.tsv -o results.blib
# With TSV report
osprey -i sample.mzML -l library.tsv -o results.blib --report results.tsv
# Custom fragment tolerance (e.g., 20 ppm)
osprey -i sample.mzML -l library.tsv -o results.blib --fragment-tolerance 20
# Unit resolution mode (uses Th/Dalton tolerances instead of ppm)
osprey -i sample.mzML -l library.tsv -o results.blib --resolution unit
# Filter output at precursor-level FDR only (less conservative than default)
osprey -i sample.mzML -l library.tsv -o results.blib --fdr-level precursor
# Protein-level FDR always runs at 0.01 by default. Change threshold or
# use razor peptide assignment:
osprey -i *.mzML -l library.tsv -o results.blib --protein-fdr 0.05 --shared-peptides razor
# Filter blib output to only peptides from protein groups passing protein FDR:
osprey -i *.mzML -l library.tsv -o results.blib --fdr-level protein
# Trace a specific peptide's journey through the pipeline (zero overhead when unset)
OSPREY_TRACE_PEPTIDE=PEPTIDEK osprey -i *.mzML -l library.tsv -o results.blib --verbose 2>&1 | tee trace.log
grep '\[trace\]' trace.log
# Use decoys already present in the library (DIA-NN / EncyclopeDIA / Carafe
# output) instead of generating them. Default prefixes recognised on protein
# accessions: DECOY_, rev_, decoy_ (case-insensitive); customise via the
# `decoy_prefixes` list in YAML. Osprey pairs each decoy with its target by
# matching amino-acid composition within a target/decoy protein pair, and
# bails if fewer than 80% of decoys pair successfully.
osprey -i *.mzML -l library_with_decoys.tsv -o results.blib --decoys-in-library
# Same, but with an explicit FDRBench pairing manifest (preferred when the
# library was generated by FDRBench). Faster, exact, and works even for
# decoy strategies that don't preserve AA composition.
osprey -i *.mzML -l library_with_decoys.tsv -o results.blib \
--decoys-in-library --decoy-pairing-manifest entrapment_pep.txt
# Emit an FDRBench-compatible input TSV alongside the blib for FDR-quality
# evaluation via entrapment counting. Includes every scored target with its
# raw SVM discriminant (not just FDR-passing entries). The output's level
# matches --fdr-level. Run FDRBench with `-score 'score:1'`.
osprey -i *.mzML -l library.tsv -o results.blib --fdrbench fdp_input.tsv# Generate a template configuration
osprey --generate-config my_analysis.yaml
# Edit the configuration file, then run
osprey --config my_analysis.yaml
# Override specific settings
osprey --config my_analysis.yaml --run-fdr 0.005input_files:
- sample1.mzML
- sample2.mzML
library_source:
DiannTsv: library.tsv
# Or: Blib: library.blib
# Or: Elib: library.elib
output_blib: results.blib
output_report: results.tsv
# RT Calibration
rt_calibration:
enabled: true
loess_bandwidth: 0.3 # Fraction of data for local fits (0.2-0.5)
min_calibration_points: 200
rt_tolerance_factor: 3.0 # Multiplier for residual SD
fallback_rt_tolerance: 2.0 # Used if calibration fails
# FDR control
run_fdr: 0.01
experiment_fdr: 0.01
decoy_method: Reverse # Options: Reverse, Shuffle, FromLibrary
decoys_in_library: false # Set true (or decoy_method: FromLibrary) to trust library decoys
decoy_prefixes: # Protein-accession prefixes that mark library decoys
- DECOY_ # (case-insensitive). Used only when decoys_in_library is true.
- rev_
- decoy_
decoy_pairing_manifest: ~ # Optional FDRBench *_pep.txt manifest for exact target-decoy
# pairing. If unset, Osprey pairs by AA composition within
# target/decoy protein pairs.
decoy_pair_min_fraction: 0.80 # Hard error if pairing succeeds for fewer than this fraction
fdr_level: Precursor # Output filtering level: Precursor (default), Peptide, Protein, or Both
# Protein-level FDR (always runs)
protein_fdr: 0.01 # Protein-level FDR threshold (default 0.01)
shared_peptides: All # How to handle shared peptides: All (default), Razor, or Unique
# Optional FDRBench-compatible input TSV for FDR-quality evaluation
output_fdrbench: ~ # Path to write FDRBench input TSV (null = disabled)
fdrbench_per_run: false # Emit one row per (precursor, run) instead of per precursorOptions:
-c, --config <CONFIG>
Configuration file (YAML format)
--generate-config <FILE>
Generate a template configuration file
-i, --input <INPUT>...
Input mzML file(s)
-l, --library <LIBRARY>
Spectral library file (.tsv for DIA-NN, .blib, or .elib)
-o, --output <OUTPUT>
Output results file (.blib format for Skyline)
--resolution <MODE>
Resolution mode: unit, hram, auto [default: auto]
- auto: Use config defaults (ppm tolerances)
- unit: Use Th (Dalton) tolerances for low-res data
- hram: Use ppm tolerances for high-res data
--fragment-tolerance <VALUE>
Fragment m/z tolerance (e.g., 10 for 10 ppm, or 0.3 for 0.3 Th)
--fragment-unit <UNIT>
Fragment tolerance unit: ppm, mz (Thompson)
--precursor-tolerance <VALUE>
Precursor m/z tolerance (e.g., 10 for 10 ppm, or 1.0 for 1.0 Th)
--precursor-unit <UNIT>
Precursor tolerance unit: ppm, mz (Thompson)
--rt-tolerance <MINUTES>
RT tolerance in minutes (fallback when calibration disabled) [default: 2]
--no-rt-calibration
Disable RT calibration (use fixed rt_tolerance instead)
--run-fdr <THRESHOLD>
Run-level FDR threshold [default: 0.01]
--experiment-fdr <THRESHOLD>
Experiment-level FDR threshold (for multi-file analyses) [default: 0.01]
--fdr-method <METHOD>
FDR method: percolator (built-in SVM, default), mokapot (external Python),
or simple (no ML) [default: percolator]
--fdr-level <LEVEL>
FDR filtering level for output: precursor, peptide, protein, or both [default: precursor]
- precursor: filter on precursor-level q-values (peptide + charge)
- peptide: filter on peptide-level q-values (peptide sequence only)
- protein: filter on protein-level q-values
- both: filter on max(precursor, peptide) q-values (most conservative)
Note: the blib output always enforces precursor-level FDR within each
eligible peptide regardless of this setting; --fdr-level controls which
peptide identities are admitted.
--protein-fdr <THRESHOLD>
Protein-level FDR threshold [default: 0.01]. Protein parsimony and
picked-protein FDR always run; this sets the cutoff for reporting
and for the compaction/reconciliation rescue rule. Use --fdr-level
protein to restrict the blib output to protein-FDR-passing groups.
--shared-peptides <MODE>
How to handle peptides mapping to multiple protein groups: all, razor,
or unique [default: all]
- all: shared peptides contribute to all their protein groups
- razor: shared peptides assigned to the group with most unique peptides
- unique: only unique peptides used; shared peptides excluded
--decoys-in-library
Trust decoys already present in the spectral library instead of
generating them. Library entries whose protein accession matches one
of the configured `decoy_prefixes` (default: DECOY_, rev_, decoy_,
case-insensitive) are flagged as decoys; the DecoyGenerator step is
skipped. Osprey then pairs each decoy with its target so their
base_ids match (required for SVM/LDA target-decoy competition).
Pairing prefers an explicit manifest via --decoy-pairing-manifest;
otherwise it falls back to matching amino-acid composition within
target/decoy protein pairs. Bails out with a clear error if no
entries match any prefix or if pairing succeeds for fewer than 80%
of decoys.
--decoy-pairing-manifest <PATH>
Path to a FDRBench pairing manifest (5-column TSV: sequence, decoy,
proteins, peptide_type, peptide_pair_index). Used with
--decoys-in-library to link each library decoy to its target
deterministically. Recommended for libraries generated by FDRBench.
Without a manifest, Osprey falls back to composition-based pairing.
--fdrbench <PATH>
Write an FDRBench-compatible input TSV. Emits every scored target
(not just FDR-passing entries) with raw SVM discriminant as `score`,
so FDRBench can compute true-FDR via entrapment counting without
truncation at Osprey's threshold. Level is auto-selected from
--fdr-level (`both` emits precursor-level; `protein` requires
--protein-fdr and writes the picked-protein TSV). Invoke FDRBench
with `-score 'score:1'`.
--fdrbench-per-run
With --fdrbench: emit one row per (precursor, run) using run-level
q-values (adds a `run` column). Default is one row per precursor
using experiment-level q-values. Ignored for protein-level output.
--write-pin
Write PIN files for external tools
--no-prefilter
Disable the coelution signal pre-filter
--no-join
HPC scoring split: run Stages 1-4 (per-file scoring) and exit
before FDR / reconciliation / blib output. Per-file
`<stem>.scores.parquet` is written next to each input mzML.
Mutually exclusive with --join-only; cannot be combined with
--input-scores.
--join-only
HPC scoring split: skip Stages 1-4 and run FDR, reconciliation,
and blib output starting from existing `.scores.parquet` caches
provided via --input-scores. Requires --library and --output.
Validates parquet footer hashes against the current config and
aborts before any work if they don't match. Mutually exclusive
with --no-join; cannot be combined with --input.
--input-scores <PATH...>
One or more `.scores.parquet` files, or a single directory
scanned non-recursively for `*.scores.parquet`. Required with
--join-only; ignored otherwise.
--parquet-compression <CODEC>
Compression codec for `.scores.parquet` writes: zstd (default,
smaller files) or snappy. Reading auto-dispatches on per-column-
chunk metadata. Use `snappy` for cross-implementation
compatibility with OspreySharp (Parquet.Net 3.x supports Snappy
only, not zstd).
--threads <N>
Number of threads (default: all available)
--report <FILE>
Write TSV report to this file
-v, --verbose
Verbose output
-h, --help
Print help (see a summary with '-h')
-V, --version
Print version
For large experiments, Osprey can split scoring (per-file, embarrassingly parallel) from joining (one process that runs FDR, reconciliation, and writes the blib):
# Score each file independently on a compute node; writes <stem>.scores.parquet next to each mzML
osprey --no-join -i file_001.mzML -l library.tsv --resolution hram
osprey --no-join -i file_002.mzML -l library.tsv --resolution hram
# ... fan out across N nodes ...
# Once all per-file parquets are written, run the join step on the head node
osprey --join-only --input-scores /path/to/parquet/dir \
-l library.tsv -o results.blib --resolution hram--input-scores accepts either a directory (scanned non-recursively for *.scores.parquet) or an explicit list of files. The join step validates each parquet's footer hashes (search parameters + library identity + binary version) before doing any work; mismatch aborts with a clear, file-named error.
For cross-implementation interop with OspreySharp, add --parquet-compression snappy on the scoring step so the resulting parquets can be read by either the Rust or the C# join.
The primary output is a SQLite database in BiblioSpec format, compatible with Skyline. It includes:
- RefSpectra: Detected peptides with consensus spectra
- RefSpectraPeaks: Fragment m/z and intensities
- Modifications: Modification positions and masses
- Proteins/RefSpectraProteins: Protein accessions and mappings
- RetentionTimes: Per-file peak boundaries with nullable retentionTime for Skyline ID lines
- OspreyPeakBoundaries: Peak boundaries (StartRT, EndRT, ApexRT) per run
- OspreyRunScores: Run-level q-values and scores
- OspreyExperimentScores: Experiment-level q-values
Import directly into Skyline for quantification. See BiblioSpec Output Schema for complete schema documentation.
Optional human-readable report with peptide detections, scores, and peak boundaries.
Each file gets independent calibration because LC conditions may vary between files:
- Sample library peptides (default 100K targets, 2D stratified on a (RT × precursor m/z) grid for uniform gradient coverage)
- Assume linear relationship: library RT range ≈ mzML RT range
- Wide initial tolerance (20-30% of gradient range)
- Score peptides and detect peaks via co-elution search
- Record (library_RT, measured_apex_RT) pairs
- Fit LOESS calibration curve using classical Cleveland (1979) robust iterations — residuals are refreshed at each iteration so the bisquare weights progressively tighten (set
OSPREY_LOESS_CLASSICAL_ROBUST=0for the legacy single-refresh behavior) - Calculate residual statistics — tight RT tolerance (3× MAD × 1.4826) used for the main search
Calibration results are saved to JSON per file for reuse on subsequent runs. See Calibration for the full algorithm including two-pass refinement.
Candidate selection uses:
- Isolation window: From mzML file (defines which precursors are fragmented)
- RT tolerance: Calibrated (Phase 2) or fallback (if calibration disabled/fails)
Note: No additional precursor m/z tolerance is applied - the isolation window from the mzML is sufficient.
Following the pyXcorrDIA approach:
-
Enzyme-aware terminal preservation:
- Trypsin/Lys-C: Preserves C-terminal residue only (the cleavage site), reverses the rest
- Lys-N/Asp-N: Preserves N-terminal residue only, reverses the rest
- Example:
PEPTIDEK(trypsin) →EDITPEPK
-
Fragment m/z recalculation:
- b-ions become y-ions at complementary positions
- y-ions become b-ions at complementary positions
- Full mass recalculation with proper ion formulas
-
Precursor mass preserved: Same amino acid composition means same precursor m/z
For each candidate precursor:
- Extract fragment XICs across the chromatographic dimension
- Compute pairwise fragment co-elution correlations
- Detect candidate peaks via CWT consensus (Mexican Hat wavelet across transitions)
- Rank candidates by
coelution_score × rt_penalty × log(1 + apex_intensity):coelution_score: mean pairwise Pearson correlation of fragment XICs within the peakrt_penalty: Gaussian centered on calibration-predicted RT, sigma = 5 × MAD × 1.4826 (downweights peaks far from expected RT without eliminating them)log(1 + apex_intensity): intensity tiebreaker so the main peak beats its own shoulder when coelution scores are comparable
- Score the selected peak using spectral matching (dot product, XCorr, hyperscore) at the apex
- Extract 21 features per precursor for machine learning scoring via built-in Percolator SVM (or optionally Mokapot)
For multi-file experiments, Osprey reconciles peak integration boundaries across replicates after the initial FDR pass:
- Collect peptides passing run-level precursor FDR across all files (per-entry precursor q-value is a hard precondition; protein FDR can upgrade borderline peptide-level evidence but cannot override poor precursor evidence)
- Compute consensus library RTs using weighted median of per-run detections, weighted by
sigmoid(SVM score)so wrong-peak detections with negative scores are downweighted - Refit per-run LOESS calibration using consensus peptides
- Derive
rt_tolerancefrom the global median of per-peptide library-RT MADs (within-peptide reproducibility floor, typically 3-5x tighter than cross-peptide calibration MAD) - For each entry in each run: keep the existing peak, switch to an alternate CWT candidate at the consensus RT, or perform forced integration
- Re-score reconciled entries and compute final experiment-level q-values
This ensures consistent quantification across replicates by aligning peak boundaries to the same chromatographic feature. See Cross-Run Reconciliation for algorithm details.
Osprey uses per-file Parquet caching to scale to 1000+ file experiments without running out of memory. After scoring each file, the full scored entries are written to ZSTD-compressed Parquet caches and replaced in memory with lightweight FdrEntry stubs (~80 bytes each). Heavy data (features, fragments, CWT candidates) is reloaded on-demand from disk only when needed. Peptide sequence strings are deduplicated using Arc<str> interning across files.
Each Parquet cache stores SHA-256 hashes of search parameters, library identity, and reconciliation state in its file metadata. On subsequent runs, Osprey validates these hashes and automatically invalidates stale caches when parameters change. SVM discriminant scores from Percolator are persisted in lightweight sidecar files, enabling Osprey to skip SVM training entirely on reruns with matching parameters.
See Pipeline Overview for memory architecture details and Intermediate File Formats for cache file documentation.
osprey/
├── crates/
│ ├── osprey-core/ # Core types and configuration
│ ├── osprey-io/ # File I/O (mzML, blib, libraries)
│ ├── osprey-chromatography/# Peak detection, RT calibration
│ ├── osprey-scoring/ # Feature extraction, decoy generation
│ ├── osprey-fdr/ # FDR control (Percolator SVM + optional Mokapot)
│ ├── osprey-ml/ # Machine learning (SVM, PEP estimation)
│ └── osprey/ # CLI and pipeline
- osprey-core:
LibraryEntry,Spectrum,OspreyConfig,IsolationWindow,FdrEntry - osprey-io:
MzmlReader,DiannTsvLoader,ElibLoader,BlibLoader,BlibWriter - osprey-chromatography:
PeakDetector,RTCalibrator,RTStratifiedSampler - osprey-scoring:
DecoyGenerator,FeatureExtractor - osprey-fdr:
FdrController,MokapotRunner - osprey-ml:
LinearSvmClassifier,PepEstimator(peptide-level PEP only; protein PEP is intentionally not computed — seedocs/07-fdr-control.md)
A Makefile provides convenience targets that run formatting and linting before each step:
make check # Format and lint (cargo fmt + clippy)
make test # Format, lint, and run tests
make build # Format, lint, and build release
make install # Format, lint, build release, and installcargo doc --open- mzML parsing via mzdata crate
- DIA-NN TSV library loading
- EncyclopeDIA elib library loading
- BiblioSpec blib library loading
- Fragment XIC co-elution analysis with ppm-based fragment matching
- CWT consensus peak detection (Mexican Hat wavelet, median across transitions)
- RT-penalized, intensity-weighted peak selection: score = coelution × Gaussian RT penalty × log(1 + apex_intensity); interferers at the wrong RT lose via the RT penalty, narrow low-intensity shoulders lose via the log-intensity factor
- Tukey median polish for robust peak boundaries and fragment scoring
- Signal pre-filter for ~30% speedup on HRAM data (disable with
--no-prefilter) - RT calibration with LOESS regression (classical Cleveland 1979 robust iterations) and (RT × precursor m/z) stratified sampling
- MS1/MS2 mass calibration
- Enzyme-aware decoy generation with fragment recalculation
- 21-feature extraction per precursor for machine learning scoring
- Built-in Percolator-style semi-supervised SVM for FDR control (no Python required)
- Two-level FDR control (run + experiment level) with dual precursor + peptide level FDR
- Two-stage blib output gate:
--fdr-leveldetermines eligible peptide identities; precursor-level FDR is always enforced within each eligible peptide (fallback to best charge state if no charge passes) - Native protein parsimony with true picked-protein FDR (Savitski 2015), two-pass architecture, and protein-aware compaction + reconciliation consensus rescue
- Optional Mokapot integration with parallel workers and progress streaming
- Cross-run peak reconciliation: aligns integration boundaries across replicates using consensus RTs; sigma-clipped MAD + original-calibration cap prevent tolerance inflation from wrong-peak detections
- Multi-file observation propagation: all per-file observations for passing precursors included in output
- Disk-backed memory architecture with per-file Parquet caching for 1000+ file experiments
- Intelligent cache invalidation via SHA-256 parameter hashing in Parquet metadata
- FDR score sidecar caching — skips Percolator SVM training on reruns with same parameters
- Streaming blib output via lightweight plan entries — avoids loading full entries into memory
- Binary spectra cache for faster second-pass mzML loading
- Cross-implementation bisection diagnostics for validation against OspreySharp (env-var gated; zero overhead when unused)
- Fragment co-elution scoring
- Mass accuracy features (ppm-level)
- MS1 isotope envelope and precursor co-elution features
- blib output with Osprey extension tables (library theoretical fragments)
- YAML configuration
- CLI with all core options
- Calibration JSON save/load and HTML report generation
- EMG peak fitting (Levenberg-Marquardt)
- Feature weight optimization (remove low-value features)
- Background correction
- Two-step search strategy
- Iterative candidate expansion
If you use Osprey in your research, please cite:
[Citation pending publication]
Apache 2.0
Developed by the MacCoss Lab at the University of Washington.
- Skyline - Targeted mass spectrometry environment
- DIA-NN - DIA data analysis
- EncyclopeDIA - Library searching for DIA
- pyXcorrDIA - Python DIA analysis tool
- mokapot - Semi-supervised FDR control