Skip to content

DIA-NN 1.9.2

Latest
Latest
Compare
Choose a tag to compare
@vdemichev vdemichev released this 21 Oct 08:03
· 3 commits to master since this release
1.9.2
af0e13d
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Learn about vigilant mode.

DIA-NN 1.9.2 is a major update with several key performance and functionality improvements.

Notes

  • DIA-NN 1.9.2 is a free Academia-only version.
  • Our spin-off Aptila Biotech is now preparing an enterprise version of DIA-NN for Industry.

Identification performance

  • Major phosphoproteomics improvement.
  • Redesigned neural network classifier with on average better performance.
  • Completely redesigned and improved mass calibration, in particular on Orbitrap and Astral instruments. The algorithm is highly effective but we know how to improve it further in future DIA-NN versions.

Quantification performance

  • Major improvement of protein quantification with QuantUMS.
  • The normalisation algorithm has been changed. It is now more reliant on the majority of the proteins being unchanged between samples but yields significantly higher precision and more proteins differentially expressed in most cases.
  • Improved quantification precision on timsTOF when using MBR.

Speed and memory

  • Improved ultra-fast mode. Combined with MBR it can now yield near-optimal performance on some phosphoproteomics datasets acquired on Orbitrap/Astral or timsTOF instruments, while providing a several-fold speedup.
  • Up to several-fold faster analysis of blanks/failed runs.
  • More than twice reduction of memory usage for the internal representation of the spectral library, this is relevant for large libraries, e.g. for phospho. Library RAM usage will be further reduced in future versions of DIA-NN.
  • Better control of memory consumption during search with large libraries.

FDR control

  • New 'Conservative' machine learning mode (experimental), which imposes the theoretical upper bound of a factor of 2 on the possible q-value deflation due to ML overfitting, if any. The mode is meant to be used with MBR.
  • The --nn-fold 4 option (experimental) that ensures that each neural network in an ensemble is only used for prediction on samples it has not been trained on.
  • Of note, these functions are normally not needed on 99% of datasets, however if the purpose is to benchmark the software and too-conservative q-values are, due to the design of the experiment, preferable to optimistic q-values, then using these options is recommended.

Usability improvements

  • Online Skyline installations support. An Administrative install of Skyline is still necessary, but DIA-NN will use it to find and launch the online install, if available.
  • Ability to run multiple Viewer instances to compare different peptides or runs side-by-side.
  • A fragment ion coverage plot is added to the Viewer. This is to be used for quick visual reference only, for making meaningful conclusions please rely directly on the extracted chromatograms shown rather than on this plot.
  • The name of an in silico predicted library to be generated is shown in the GUI with the correct extension.

Fixes

  • The bug on Linux which manifested as a crash when using the --matrices option has been fixed.
  • The bug that caused incorrect results when using on-the-fly in silico prediction from FASTA combined with raw files searching in the same DIA-NN run and with peptidoform scoring enabled.

Notes

  • The documentation will be updated to describe DIA-NN 1.9.2 after HUPO.
  • An update of the Linux binary was added on October 31, 2024, fixing an issue with memory allocation (no functional changes).
Assets 6
Loading