Deep learning the collisional cross sections of the peptide universe from a million training samples

Abstract

The size and shape of peptide ions in the gas phase are an under-explored dimension for mass spectrometry-based proteomics. To explore the nature and utility of the entire peptide collisional cross section (CCS) space, we measure more than a million data points from whole-proteome digests of five organisms with trapped ion mobility spectrometry (TIMS) and parallel accumulation -serial fragmentation (PASEF). The scale and precision (CV <1%) of our data is sufficient to train a deep recurrent neural network that accurately predicts CCS values solely based on the peptide sequence. Cross section predictions for the synthetic ProteomeTools library validate the model within a 1.3% median relative error (R > 0.99). Hydrophobicity, position of prolines and histidines are main determinants of the cross sections in addition to sequence-specific interactions. CCS values can now be predicted for any peptide and organism, forming a basis for advanced proteomics workflows that make full use of the additional information.

Ultra-high sensitivity mass spectrometry quantifies single-cell proteome changes upon perturbation

“…3C). To achieve high data completeness between single-cell measurements, we next replaced dda by diaPASEF, in which fragment level matching is further supported by ion mobility data 35 . We found that combining diaPASEF scan repetitions further improved protein identification numbers (Fig.…”

“…Chemical noise is widely distributed as a result of its heterogeneous nature and the ten-fold increased peak capacity due to TIMS (Fig. 1a, b) 24 . These precursors can be fragmented in a highly sensitive manner, either in data dependent (ddaPASEF) or data independent (diaPASEF) mode, resulting in very high ion utilization and data completeness 25 .…”

Sequence-specific model for predicting peptide collision cross-section values in proteomic ion mobility spectrometry

“…At the time of our work, Meier et al concurrently developed a deep learning CCS prediction model using 570,000 unique combinations of sequence, charge state (2+, 3+ and 4+), including peptides with oxidized methionine. 38 However, many machine learning approaches often operate as "black boxes", providing limited information on the underlying separation mechanisms. Meier et al have demonstrated the contributions of the 20 amino acid residues and the qualitative trends for hydrophobicity, Pro content, and position of His.…”

“…Meier et al have demonstrated the contributions of the 20 amino acid residues and the qualitative trends for hydrophobicity, Pro content, and position of His. 38 They highlighted the difficulty to model the observed physicochemical properties along with sequence dependent features directly with the linear sequences and our work here is able to address such difficulties as well as investigate finer composition and position dependent features that are novel to our approach. Therefore, a semiempirical Sequence-Specific Retention Calculator (SSRCalc) approach based on positiondependent correction coefficients was applied in this work to establish a Sequence-Specific Ion mobility Calculator (SSICalc).…”

DreamDIA-XMBD: deep representation features improve the analysis of data-independent acquisition proteomics

“…Numeric results have confirmed that for these data, features from deep representation networks outperform those from conventional feature engineering strategies as the latter may not be able to exploit the intrinsic joint distribution of the signals in the high-dimension feature space towards the designated target of learning 35 . For DIA data analysis 36 , deep learning has also been recently introduced for the prediction of fragment intensities, retention time and ion mobility [37][38][39][40][41] , and de novo sequencing 42 .…”