Search bioRxiv⌕ Search

Biology subjects

Tiwary, S.

Publications and source records attributed to Tiwary, S..

2 recordsLinked to original sources
bioRxiv2024

Tesorai Search: Large pretrained model boosts identifications in mass spectrometry proteomics without the need for Percolator.

The original mass spectrometry search engines used simple algorithms for peptide identification. Recent tools improved accuracy by adding several extra components such as fragment ion intensities or retention times prediction and training target-decoy classifiers on-the-fly, leading to sometimes inconsistent results. Our study explores the impact of replacing those extra components with a deep-learning pretrained model that directly learns the complex relationship between the full spectra and associated peptide sequence, without using decoys. This simplified workflow has fewer parameters to tweak, making it easier to use and perform robustly on data from instruments and use-cases never seen during training. Surprisingly, our approach consistently identifies more peptides than FragPipe, PEAKS, and Proteome Discoverer (12%, 9%, and 21% more, respectively, across a range of datasets). Tesorai Search is also fast - 250 immunopeptidomics searches in 45 minutes - and free for academics, available as a webserver at console.tesorai.com.

bioinformatics↗
bioRxiv2024

CoSpred: Machine learning workflow to predict tandem mass spectrum in proteomics

In mass spectrometry-based proteomics, the identification and quantification of peptides and proteins is usually done using database search algorithms or spectral library matching. The use of deep learning algorithms can help improve the identification rates of peptides and proteins through the generation of high-fidelity theoretical spectrum which can be used as the basis of a more complete spectral library than those presently available. Current methods focus on predicting only backbone ions, such as y- and b-ions. However, the inclusion of non-backbone ions is necessary to truly improve spectral library matching. Here we focus on providing a user-friendly machine learning workflow, which we call Complete Spectrum Predictor (CoSpred). Using CoSpred users can create their own machine learning compatible training dataset and then train a Machine Learning model to predict both backbone and non-backbone ions. For the model a transformer encoder architecture is used to predict the complete MS/MS spectrum from a given peptide sequence. This model does not require background knowledge of fragment ion annotations or fragmentation rules. The model outputs the set of pairs (Mi, Ii) where Mi is the m/z (mass-to-charge ratio) of a peak in the spectrum and Ii is the intensity of the peak. The model presented here for validation was trained on the dataset available in the MassIVE data repository and shows superior performance in terms of various metrics (e.g. precision/recall for mass, cosine similarity for peak intensity, etc) between the true and predicted spectra. Furthermore, CoSpred can be used to create custom models that allow for accurate spectrum prediction for different experimental conditions. In addition to the transformer model provided in the package, the code is built modularly to allow for alternate ML models to be easily "plugged in". The CoSpred workflow (preprocessing->training->inference) provides a path for state-of-art ML capabilities to be more accessible to proteomics scientists.

bioinformatics↗