Accounting for digestion enzyme bias in Casanovo

Abstract

A key parameter of any proteomics mass spectrometry experiment is the identity of the enzyme that is used to digest proteins in the sample into peptides. The Casanovo de novo sequencing model was trained using data that was generated with trypsin digestion; consequently, the model prefers to predict peptides that end with the amino acids "K" or "R." This bias is desirable when the Casanovo is used to analyze data that was also generated using trypsin but can be problematic if the data was generated using some other digestion enzyme. In this work, we modify Casanovo to take as input the identify of the digestion enzyme, alongside each observed spectrum. We then train Casanovo with data generated using several different restriction enzymes, and we demonstrate that the resulting model successfully learns to capture enzyme-specific behavior. However, we find, surprisingly, that this new model does not yield a significant improvement in sequencing accuracy relative to a model trained without the enzyme information but using the same training set. This observation may have important implications for future attempts to make use of experimental metadata in de novo sequencing models.

A transformer model forde novosequencing of data-independent acquisition mass spectrometry data

“…We anticipate a number of future directions for work to further improve Cascadia on DIA data. First, Cascadia can be specifically fine-tuned for specific applications of interest, for example by training on a dataset of MHC bound peptides to improve immunoproteomics analysis [28] or non-tryptic data to ameliorate the tryptic bias of the model [29]. Furthermore, there are many successful ideas proposed by de novo methods in the DDA setting which may also be beneficial in the DIA setting, including the addition of additional auxiliary training tasks [4,9], alternate decoding strategies [6,15], and post-processing algorithms to refine predictions [19,30].…”