CHIMERYS: An AI-Driven Leap Forward in Peptide Identification

Significance of the Topic

Comprehensive identification of peptides is fundamental in bottom-up proteomics. Chimeric spectra, where multiple peptides cofragment, account for over forty percent of data from data dependent acquisition and hinder spectrum matching. Advanced algorithms are needed to fully leverage spectral complexity and improve proteomic depth.

Objectives and Study Overview

This work introduces CHIMERYS, an artificial intelligence powered search algorithm designed from the ground up to deconvolute complex tandem mass spectra and boost peptide identifications. The study benchmarks CHIMERYS against standard Sequest HT workflows across diverse sample types and acquisition parameters. Key aims include evaluation of identification rates, FDR calibration, quantification accuracy, and acquisition efficiency.

Methodology

Comparative analyses were performed using HeLa cell digests acquired on various Thermo Scientific mass spectrometers with different gradient lengths and isolation windows. Public datasets from human, yeast, plant, and body fluid samples were reanalysed. An entrapment strategy appended plant databases to human sequences for independent FDR assessment and an in silico chimeric spectra system assessed sensitivity. Search workflows in Proteome Discoverer 3.0 employed Sequest HT, INFERYS rescoring and CHIMERYS. Peptide fragment intensities and retention times were predicted by the deep learning framework INFERYS 2.0 and refined via data driven learning. All candidates within each MS2 isolation window were scored concurrently under PSM level FDR control by Percolator.

Instrumentation Used

• Thermo Scientific Orbitrap Exploris 480 mass spectrometer
• Thermo Scientific Orbitrap Eclipse Tribrid mass spectrometer
• Thermo Scientific Vanquish Neo UHPLC system
• Proteome Discoverer 3.0 software with CHIMERYS, INFERYS and Percolator nodes

Main Results and Discussion

CHIMERYS doubled peptide identifications compared to Sequest HT in classical data dependent acquisition, achieving identification rates above eighty percent and an average of two PSMs per spectrum. Peptide identifications per protein increased by approximately 2.5 fold, enhancing sequence coverage and quantification of low abundant peptides. Entrapment analyses confirmed well calibrated FDR estimation, and dilution series validated accurate quantification with 1.8 fold more proteins quantified. Simulated chimeric spectra experiments demonstrated sensitivity above ninety one percent. Wider isolation windows and shorter gradients yielded higher PSM throughput without loss of identification quality.

Benefits and Practical Applications of the Method

• Increased proteome coverage in single shot experiments
• Improved label free quantification accuracy
• Enhanced throughput via shorter chromatography gradients
• Resilience to highly complex samples across biological kingdoms
• Integrated workflow within existing software environment

Future Trends and Opportunities

Integration of intelligent deconvolution approaches promises further increases in throughput as instrument sensitivity improves. Extension to multiplexed quantification, real time data acquisition optimization, and multi-omics integration are anticipated. Wider adoption in quality control and clinical proteomics may enable robust high throughput assays. Continued refinement of deep learning based spectral prediction will drive identification depth.

Conclusion

CHIMERYS represents a paradigm shift in peptide identification by leveraging artificial intelligence for simultaneous deconvolution of chimeric spectra. The algorithm substantially improves identification rates, quantification accuracy, and acquisition efficiency. Its cloud native, microservice architecture and integration into established workflows facilitate immediate gains in proteomic studies.

References

• Zolg DP et al Rapid Commun Mass Spectrom 2021; e9128. https://doi.org/10.1002/rcm.9128
• Dorfer V et al J Proteome Res 2018;17(8):2581-2589. https://doi.org/10.1021/acs.jproteome.7b00836
• The M et al J Am Soc Mass Spectrom 2016;27(11):1719-1727
• Muller JB et al Nature 2020;582:592-596. https://doi.org/10.1038/s41586-020-2402-x
• Bian Y et al Nat Commun 2020;11:157. https://doi.org/10.1038/s41467-019-13973-x