A CCS-centric HLA-specific trained de novo module for precise and accurate real-time immunopeptide identification on the Bruker ProteoScape platform

Since its conception, Bruker ProteoScape (BPS) has transformed into a comprehensive proteomics data analysis platform that can integrate third-party tools while utilizing the concept of data streaming to realize fully customizable real-time processing workflows. To expand the capabilities of the BPS platform for immunopeptidomics, and other applications, we previously developed and integrated a timsTOF optimized de novo sequencing engine from Rapid Novor Inc., called BPS Novor. To further develop this tool and support the extremely rapidly growing field of timsTOF based immunopeptidomics, we have retrained this module with >1.4 million spectra mapping to >150,000 HLA-presented peptides from MHCI and MHC II classes.

Methods & Results:

Novor was re-trained on a variety of timsTOF acquired data, where ground truth is taken from the ProLuCID database search results filtered to 1% PSM FDR. Previously we have shown that BPS Novor is highly accurate and precise on a variety of datasets. On amino acid level, at 75% precision, BPS Novor achieved between 40-60%, whereas standard Novor achieved between 25-50%. We have also shown that BPS Novor is extremely fast by evaluating the processing speed across 5 datasets, with an average processing speed of 1338±226 spectra/second.

Here we compare the re-trained BPS Novor (MHC model) across two immunopeptidomics datasets using standardized numbers of computing cores. We focused on two recently published immunopeptidomics datasets (Feola et al., 2021 and Phulphagar et. al., 2023) to show the newly optimized BPS Novor MHC model increases precision and accuracy over the previous versions for immunopeptidomics applications, while retaining the processing speed advantage over other algorithms.

We show in the Feola et al., dataset modest improvement of 2% in amino acid correctness and 1% at the peptide correctness, with a marginal increase in processing time. For the 10 melanoma derived samples from Phulphagar et. al. A ~6% increase in amino acid and peptide correctness was observed. The MHC model showed on average 8.5% increase in the number of 8-11mer peptide sequences in the dataset.

Together, these datasets confirm that the MHC optimized scoring model for BPS Novor provides greater accuracy for immunopeptidomics data.

Conclusion

We here introduce a purposefully optimized BPS Novor module, providing on-the-fly real-time de novo sequencing for timsTOF immunopeptidomics data.

Conflict of Interest Disclosure

R.Z., Q.L., M.X., B.M. are employees of Rapid Novor, Inc. Q.L., M.X., B.M. are co-founders of Rapid Novor, Inc. D.T., T.S., J.K., G.R. are employees of subsidiaries of Bruker Corp. Novor is a product of Rapid Novor, Inc. BPS Novor is a product of Rapid Novor, Inc. sold and distributed by subsidiaries of Bruker Corp.