Faster and more accurate intensity-based PTM localization in Chimerys

Background: Acetylation and ubiquitination dynamically regulate protein function and lifespan. Mass spectrometry-based bottom-up proteomics excels at detecting and quantifying modified peptides. Initially, data dependent acquisition (DDA) was often used due to its simplicity and precise precursor isolation. However, data independent acquisition (DIA) is gaining momentum despite challenges in post-translational modification (PTM) identification and localization. We integrated intensity-based PTM localization for DDA, WWA and DIA data into Chimerys and revamped its architecture to substantially reduce DIA data processing time.

Methods: Chimerys, a spectrum-centric and acquisition method-agnostic search algorithm, deconvolutes chimeric spectra by maximizing the amount of explained experimental intensity with a minimal peptide set. Predictions stem from Inferys, a deep learning framework, formerly limited to unmodified, oxidized, and phosphorylated peptides. To support acetylation and ubiquitination, we included 16 M PSMs with common PTMs from public repositories in our training data foundation, crafting a new deep learning model. Integrated into our algorithm, this model powers an intensity-based PTM localization module, and an architecture overhaul improves CHIMERYS" DIA data processing speed. This version is integrated into Thermo Scientific™ Proteome Discoverer™ 3.2.

Results: Chimerys' capabilities depend on the peptide tandem mass spectra our model was trained on. To surmount this, we deployed a cloud-native pipeline to access and process PRIDE datasets, merging them with our training data. The updated model now supports various PTMs, like N-terminal and lysine acetylation, lysine ubiquitination, TMT0 & TMTpro 'zero' and 'super heavy', unmodified cysteine residues, and the PreOmics™ iST-NHS cysteine modification. With extra training data, prediction accuracy increased, notably for phosphorylated peptides. Leveraging these advancements, we reanalyzed a public dataset, yielding a ~30% increase in identified modified peptides versus Sequest HT. Moreover, we introduced an innovative intensity-based algorithm for PTM localization. It compares predicted spectra of all potential isomers against experimental data, furnishing a localization score for each modified residue. This algorithm surpasses ptmRS on DDA datasets and adeptly localizes modified residues even in complex DIA spectra, embracing a spectrum-centric methodology. Furthermore, to facilitate large-scale DIA experiments, we fine-tuned the CHIMERYS workflow. Individual raw files are now submitted separately, expediting job execution. Additionally, a scoring logic overhaul slashed individual DIA raw file processing time twofold. These improvements, along with a revamped workflow in Proteome Discoverer, expedite processing and scale up experiments.

Conclusion: A major update to a spectrum-centric, acquisition-method-agnostic search algorithm, enhancing PTM compatibility, and user experience.