Deep learning based co-elution scoring algorithm

In recent years, DIA technology-based mass spectrometry methods have gained extensive attention in proteomics research due to their high throughput and high sensitivity. However, the interference of the signal co-elution phenomenon in complex samples seriously affects the correct identification and quantitative analysis of peptides. In this paper, we propose a deep learning model architecture for enhancing the analysis of mass spectrometry XIC data. The model architecture consists of an input layer, a bi-directional LSTM layer, a Transformer Encoder layer, a fully-connected layer, and a Sigmoid layer, improving the identification of peptide signals by layer-wise feature extraction and global context modeling. The following pre-processing operations including chromatogram extraction, shuffle, normalization, and smooth are performed on the input data to ensure that the signal is of high quality and consistency before data entry. Experimental results verify our model, which can exactly identify target and decoy peptides, reduce the noise interference effectively, and improve the quantitative analysis reliability significantly. The method described in this paper provides an effective tool for the in-depth mining of mass spectrometry data and shows a broad application prospect in the field of complex proteome analysis.