TIMS DIA-NN 3.0: improved algorithms to enhance peptide detection with confidence and accuracy

Due to the rising popularity of data-independent acquisition (DIA), we previously developed and launched an automated spectral library generation tool and the CCS-aware TIMS DIA-NN 2.0 for DIA analysis, for the purpose of analyzing high-throughput mass spectrometry data. Subsequent to this initial launch, we have continuously put effort into creating TIMS DIA-NN 3.0 by redesigning and implementing new algorithms. The primary focus in this iteration has been on substantially improving the confidence, accuracy, and reliability of both identification and quantification processes.

In processing raw files, the TIMS DIA-NN effectively identifies local maxima by clustering and merging peaks, employing a 4D-proteomics approach. This method integrates key dimensions - collision cross-section (CCS), retention time, mass-to-charge ratio (m/z), and MS/MS fingerprinting - to optimize peak detection. Enhancements have been made with the introduction of multiple new algorithms, including new mass drift and mass tolerance calibration algorithms, which improve the detection of peptide candidates and the precision of chromatogram construction. Additionally, we have integrated novel algorithms, including a Gaussian correlation score for peak detection in neural network analysis, peak boundary detection in chromatograms, and fragment ion selection for MBR to improve reproducibility. These improvements collectively contribute to the improvement of the consistency of the coefficient of variation (CV) across replicates, thereby elevating the robustness and accuracy of the data analysis.

We ran both TIMS DIA-NN 3.0 and 2.0 on the Hela DIA triplicate data, which was acquired from the Bruker timsTOF Pro, for benchmarking purposes. The analysis was performed using a spectral library generated from a ProteoScape DDA search. While TIMS DIA-NN 2.0 identified 62,402 peptides with a CV of 16.61, TIMS DIA-NN 3.0 identified 64,507 peptides with a reduced CV of 12.59. Furthermore, when using a standard tri-species Human/Yeast/EColi; of the total 10304 proteins identified, 9698 (94%) are quantified with a CV <20%, and 7778 (75%) are quantified with a CV <10%, indicating excellent quantitation.

The TIMS DIA-NN 3.0 has been integrated into the Bruker ProteoScape platform and is available to all timsTOF and ProteoScape users.