What’s new in Mascot Server 3.2 (pre-release draft)
Note: This is a pre-release draft and the details may change before the final release.
Mascot Server is a powerful search engine which uses mass spectrometry data to identify proteins from DNA, RNA and protein sequence databases.
Mascot DIA
Mascot Server and Mascot Distiller can now be used for identifying and quantifying proteins from Data Independent Acquisition (DIA) runs. This is enabled by two key features:
- New in Mascot Distiller 3.0: Novel peak detection for Thermo DIA and SCIEX SWATH (see What’s new in Mascot Distiller 3.0)
- New in Mascot Server 3.2: Noise-resistant scoring for DIA spectra (see below)
Mascot DIA is a spectrum-centric solution and uses the same analysis workflow for DDA and DIA identification and quantitation. Read more in Mascot DIA.
New: Noise-resistant scoring for Data Independent Acquisition (DIA)
DIA spectra are highly chimeric, containing fragments from multiple precursors as well as unfragmented, overlapping precursor signal. Mascot Server 3.2 introduces a new probabilistic scoring model that is much more resistant to ‘noise’ and peaks from unrelated precursors.
When matching fragments to observed peaks, Mascot selects peaks in intensity order within a peak selection window. The peak selection window for DDA is 100Da and for DIA, 55Da, allowing Mascot to use twice as many peaks for scoring.
The screenshot below compares sensitivity at 1% sequence FDR for a SCIEX TTOF6600 SWATH run with 99 variable isolation windows (PRIDE project PDX028375). Precursor detection and peak picking was performed by Mascot Distiller 3.0. The DDA probabilistic scoring struggles to identify just 8,806 peptide sequences from the 2h elution. With noise-resistant scoring, Mascot Server 3.2 identifies 14,534 peptide sequences.
Noise-resistant scoring is triggered by the new search parameter DATA_INDEPENDENT_ACQUISITION=1. Mascot Distiller 3.0 sets the search parameter automatically for DIA projects. Peptide scoring for DDA data is unchanged in this version of Mascot Server.
Improved: MS2Rescore 4.0, MS2PIP 4.2 and DeepLC to 4.5
MS2Rescore, MS2PIP and DeepLC have been updated to the latest version.
- DeepLC: both execution speed and prediction accuracy are greatly improved
- MS2PIP: execution speed is greatly improved
DeepLC 4 simplifies model choice: there is now just one model, “auto”, which is automatically calibrated to your search results. Prediction accuracy is dramatically improved, and DeepLC is now usable with any gradient length from 5 minutes to 2 hours.
MS2PIP shipped with Mascot Server now computes fewer features; poorly performing or redundant features have been removed, leading to quicker downstream execution and fewer issues with Percolator.
Improved: Percolator has been updated to version 3.7
Mascot Server 3.1 shipped with Percolator 3.5. In some data sets, enabling an MS2PIP or DeepLC model caused Percolator 3.5 to fail to estimate a PEP (posterior error probability), which led to all peptide matches getting a score of zero and becoming non-significant. The Percolator executable bundled with Mascot Server has been updated to Percolator 3.7, which solves the issue.
Improved: Core feature set for refining results yields 5-15% more matches
We have reviewed the core set of features calculated from PSMs, which are used in refining with machine learning. The core feature set was last reviewed in Mascot Server 2.8. On average, the new feature set yields 5-15% more peptide matches across a wide range of DDA runs compared to Mascot Server 3.1.
The previous feature set included four features that are very slow to compute: rawScore, isoSysDM, isoSysDMppm and isoSysDMz. With the new feature set, these dimensions no longer provide useful discrimination between targets and decoys, so they have been removed. This makes refining much faster, often taking half as much time as Mascot Server 3.1.
Improved: Execution speed when calculating core features is improved
Performance bottlenecks in the component that calculates core features (ms-createpip.exe) have been addressed. This step now takes approximately half the time compared to version 3.1.
Improved: Percolator multithreading speed
Percolator shipped with Mascot now uses more threads, greatly reducing the amount of time spent in rescoring.
Improved: Mascot Daemon 3.2
Mascot Daemon is the automation client shipped with Mascot Server. Mascot Server 3.2 ships with the new Mascot Daemon 3.2.
New: Refine merged results (LFQ) with machine learning [DDA, DIA]
Replicate (LFQ) quantitation using Mascot Distiller creates a merged results report, where peptide matches from different runs are combined into a single report. Mascot Server 3.2 can now refine the merged results using machine learning. The refined results can be exported in CSV and mzIdentML formats.
Refining merged results also enables machine learning when creating a quantitation summary report using the Replicate protocol in Mascot Daemon 3.2.
Mascot Server uses a train-then-merge strategy. Individual runs are refined separately, then the refined results are merged. This strategy is required when a DeepLC model is selected, because the predicted retention times are calibrated per run. The train-then-merge strategy is also used when DeepLC is not active.
Improved: Database searches are 30-50% faster
We have made several performance improvements to database searches:
- Algorithm hotspots and chokepoints have been reduced or eliminated
- Memory management has been redesigned
- Thread scheduling has been improved
Both DDA and DIA searches should be 30%-50% faster on the same hardware compared to Mascot Server 3.1.
Improved: Intel Core Ultra and Alder/Raptor/Arrow Lake processors on Windows
If you have an Intel processor with heterogeneous architecture (‘P’ and ‘E’ cores), and you run Mascot Server on Windows, there were previously issues with the operating system trying to run everything on the slow E cores. In version 3.2, nph-mascot.exe disables EcoQoS mode at the start of the database search to ensure the operating system runs threads on performance (P) cores. This gives a further performance boost to Intel Alder/Raptor/Arrow Lake processors and Intel Core Ultra processors.
Improved: Database compression speed on Windows
On some Windows systems, compressing a sequence database may take longer than the equivalent Linux version. We have changed the regular expression engine used in this step. Compressing databases on Windows should now be 2-3 times quicker.
Improved: Processing and accessing MSR results files is faster
Mascot Server 3.0 introduced the MSR format for results files. We have identified and fixed several performance issues with opening, caching, processing and accessing search results in MSR format:
- The caching step between end of database search and viewing results is 4-5 times shorter.
- Loading Report Builder and exporting results (in any format) should be at least twice as fast, sometimes an order of magnitude faster.
Additionally, caching and accessing MSR files takes half the amount of RAM compared to version 3.1.
Improved: Exporting refined results from client.pl?result_file_mime was unnecessarily slow
The API command client.pl?result_file_mime is used by Thermo Proteome Discoverer as well as Mascot Distiller 2.x. When refining with machine learning is enabled, the command now runs much faster due to removing redundant isValid() checks, up to 6 times faster than Mascot Server 3.1. The impact can be seen by comparing the time to import Mascot results into the client application.
New: Allow semi-specific enzyme with ET search
The Mascot Error Tolerant search is a two-pass search. Previously, the enzyme selected for the first pass had to be specific at both ends. You can now submit an Error Tolerant search with a semi-specific enzyme. In the second pass, Mascot iterates through all variable modifications and now allows a semi-specific peptide to carry an error tolerant modification.
For example, if you treat glycopeptides with O-Glycoprotease (IMPa), it cleaves at N-terminal serine or threonine but only if the serine/threonine is glycosylated, creating semi-tryptic peptides. You can now search identify N-glycans on these peptides by submitting an Error Tolerant search with semi-trypsin.
Improved: systemd unit file for Linux
Mascot Server 3.2 includes a new systemd unit file, which integrates the Mascot service (ms-monitor.exe) with the system job control. This replaces the old SysV-init script shipped with previous versions of Mascot.
Removal of obsolete functionality
Support for Windows 8.1 and Server 2012 R2 has been dropped. On Windows, Mascot Server 3.2 requires Windows 10 or later, or Windows Server 2016 or later. We recommend Windows 11/Server 2025.
Other improvements
The release also fixes other minor issues. In total, almost 200 change requests and bugs are addressed in this release. The full list is included in the release notes.
