To Fly, or Not to Fly, That Is the Question: A Deep Learning Model for Peptide Detectability Prediction in Mass Spectrometry

• DOI: 10.1021/acs.jproteome.4c00973
• Publication Date: 2025-05-09T18:34:01Z
• Authors: Naim Abdul-Khalek; Mario Picciani; Omar Shouman; Reinhard Wimmer; Michael Toft Over...; Mathias Wilhelm; Simon Gregersen E...
• Keywords: On the fly

Identifying detectable peptides, known as flyers, is key in mass spectrometry-based proteomics. Peptide detectability strongly related to peptide sequences and their resulting physicochemical properties. Moreover, the high variability MS data challenges development of a generic model for prediction, underlining need customizable tools. We present Pfly, deep learning developed predict based solely on sequence. Pfly versatile reliable state-of-the-art tool, offering performance, accessibility, easy customizability end-users. This adaptability allows researchers tailor specific experimental conditions, improving accuracy expanding applicability across various research fields. an encoder-decoder with attention mechanism, classifying peptides flyers or non-flyers, providing both binary categorical probabilities four distinct classes defined this study. The was initially trained synthetic library subsequently fine-tuned biological dataset mitigate bias toward synthesizability, predictive capacity outperforming predictors benchmark comparisons different human cross-species datasets. study further investigates influence protein abundance rescoring, illustrating negative impact identification due misclassification. has been integrated into DLOmix framework accessible GitHub at https://github.com/wilhelm-lab/dlomix.