DIA-NN 1.9.2

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).

DIA-NN 1.9.1

DIA-NN 1.9.1 is a minor update of 1.9.

• Linux version included
• Faster processing of large Slice-PASEF runs
• Emprical DIA-based libraries are now saved in .parquet format instead of .tsv: less disk space, precise real numbers
• Fixed a bug in the implementation of --no-cut-after-mod
• Fixed a bug in processing contaminants when doing on-the-fly FASTA digest and raw data analysis
• Default output location in the GUI is now C:/Temp, if exists, instead of the DIA-NN installation folder
• The .protein_description.tsv output file contains the protein information used by DIA-NN
• The .manifest.txt output file now provides a description of output files produced
• Adjusted settings for launching Skyline

For the future DIA-NN roadmap, see https://github.com/vdemichev/DiaNN/releases/tag/1.9

DIA-NN 1.9

DIA-NN 1.9 release summary

DIA-NN 1.9 is the biggest improvement of DIA-NN so far. Below is the summary of key features, please see the documentation for details.

Peptidoforms
Data-dependent acquisition (DDA) has so far maintained one key advantage over data-independent acquisition (DIA): confidence in peptidoform assignment. That is, with DDA one can be reasonably confident that a peptide is matched to the spectrum of the correct peptidoform (i.e. without amino acid substitutions or other modifications), and that the set of reported modifications (phosphorylation, etc) is correct. Now we achieve this also with DIA, while maintaining all the advantages of DIA and largely preserving its deep proteome coverage. We expect a range of applications, from pQTL analysis in population proteomics to metaproteomics. A preprint describing the new peptidoform-scoring module in DIA-NN is to follow.

Phosphoproteomics
We use the new peptidoform scoring module to significantly improve phosphoproteomics workflows. Moreover, DIA-NN now reports site-specific localisation confidence along with site-level quantities, in a convenient format, greatly simplifying its use for phosphoproteomics.

Multiplexing
DIA-NN 1.9 features a second-generation plexDIA (multiplexing) module, with a significantly enhanced ability to gain channel-specific confidence in peptide and protein identifications. Further, processing of multiplexed DIA data is greatly simplified by convenient output, including channel-specific protein group quantities obtained with QuantUMS.

timsTOF proteomics
DIA-NN 1.9 implements Slice-PASEF as well as features preliminary support for midia-PASEF and Synchro-PASEF.

Quantification
DIA-NN 1.9 features a second-generation QuantUMS module, wherein quantities are optimised with machine learning and statistically-justified accuracy estimates are available for individual quantities.

Visualisation
This has been the most often requested feature since the conception of DIA-NN. Now supported via either Skyline integration or via a dedicated DIA-NN Viewer.

General performance
Better identification numbers and stricter control of false discoveries, along with extensive options to tailor the identification and quantification confidence control to a specific experiment.

Speed and code quality
DIA-NN has been overhauled to match the modern coding practices using C++20, with a focus on efficient memory use and better multithreading. DIA-NN 1.9 features code optimisations which yield roughly 1.3x-2x speed gains for library-free search. Large predicted libraries (tens of millions of precursors) are now often 10x+ quicker to generate.

Timeline. This is a Windows release of DIA-NN 1.9, Linux support is to follow shortly. Further, we have a number of features and performance improvements under active development and will likely release a series of updates implementing these in the near future. We will also be grateful for any feedback on DIA-NN 1.9 as well as feature requests, which we will do our best to implement.

Future roadmap

DIA-NN is under active development, towards (i) enabling new technologies as well as (ii) achieving better performance for existing workflows. In the latter case, we have the following planned or under development:

• While DIA-NN performs remarkably well in library-free setting already, there is a room for even better performance. Specifically, DIA-NN will in the future implement experiment-specific transfer learning, similar to the concept recently introduced in AlphaDIA.

• DIA-NN already implements a low RAM usage mode, which restricts the amount of system memory it needs for its search. Currently, the biggest factor in RAM usage by DIA-NN in the lib-free mode is the storage of the predicted library in memory, especially when using multiplexing. DIA-NN will in the future implement a different format for internal library storage, with fold-change lower memory requirements.

• The ultra-fast mode in DIA-NN is great for preliminary analyses (up to 5x faster), although it does sacrifice identification performance, as it implements a spectrum centric-like search strategy, which is inherently less sensitive. We have a different fast search mode in works, which will have minimum performance trade offs.

• We have a number of algorithms in works, which will fully explore the potential (in terms of both identification and quantification performance) of Slice-PASEF, midia- and Synchro-PASEF, Scanning SWATH and Orbitrap Astral. While the current algorithms perform remarkably well already, showing the potential of these technologies, we work on specific improvements that will further boost the performance.

• DIA-NN will in the future incorporate a module for detailed QC analysis of DIA runs.

• Together with our collaborators, we are developing some exciting new workflows combining different tools.

DIA-NN 1.8.1

• Multiplexing support
• Improved dia-PASEF performance
• 'Peak height' quantification mode
• Fixed handling of cases when a spectral library annotates the precursor ion as fragment or includes y1/b1 fragments
• Stability issues under Linux solved

DIA-NN 1.8

A major improvement in terms of both performance and functionality. Some key changes:

• More peptide & protein IDs.
• Stringent control of global precursor and protein FDR, validated on thousands of samples.
• Transformatively better library-free analysis mode - in many cases you will no longer need a spectral library, even for the most challenging samples.
• Full support for PTM confidence scoring and site localisation. Validated workflows for phosphoproteomics and ubiquitinomics.
• DIA-NN is now considerably faster and requires less memory.
• Fully functional Linux builds - now with support for deep learning & native support for dia-PASEF data.

This release is also considerably better in some aspects than the beta versions we have been sharing.

Of note, we received multiple feature requests. Unfortunately, only a few could be implemented in the limited amount of time we had, as this release had to be scheduled for today due to the need to reference it in a publication.

Multiple papers describing the new DIA-NN features implemented in this version are to appear in the near future.

DIA-NN 1.7.18 development build

Update README.md

DIA-NN 1.7.17 development build

Update README.md

DIA-NN 1.7.16 development build

Significant improvement of multiple algorithms & functionality changes since 1.7.12, please contact by email (provided in the manual) for details on the new functionality. Minor improvement of identification performance in comparison to 1.7.15. An updated manual describing all the new features will be available in the future.

Lack of manual is the primary reason why this version is marked as a 'development build' and not a 'release'.

DIA-NN 1.7.15 development build

Significant improvement of multiple algorithms & functionality changes, please contact by email (provided in the manual) for details on the new functionality. An updated manual describing all the new features will be available in the future.

DIA-NN 1.7.14 development build

Significant improvement of multiple algorithms & functionality changes, please contact by email (provided in the manual) for details on the new functionality. An updated manual describing all the new features will be available in the future.