TMT-Labeled Annotated Spectra Prediction using Deep Neural Network

Peptides offer a promising avenue for the treatment and prevention of cancer by enabling highly targeted, effective, and safe approaches. Identification of peptides is mainly done through database searching and spectral library matching methods. However, database size and spectrum complexity respectively limit the effectiveness of peptide identification. The competency to precisely predict spectra from peptide sequences can strengthen peptide identification and protein quantification in proteomics. This work introduces a Deep Neural Network (DNN) designed to predict Tandem Mass Tag (TMT)-labeled annotated spectra to support peptide identification experiments in proteomics. The proposed DNN model was trained on the COREAD (COloREctal cAncer cell lines Data) experimental datasets. The model is trained on the annotated spectra in which TMT peaks are preserved. It incorporates TMT modifications, allowing them to predict spectra with annotated peptides. The architecture uses a sequence-to-sequence model based on a Convolution Recurrent Neural Network with Attention mechanism (CRNNA), capturing correlations between peptide sequences and their spectra. The model's performance is evaluated using cosine similarity between predicted and experimental spectra, showing remarkable performance in spectrum prediction to identify peptides that were not identified earlier from the data. The study emphasizes the utility of machine learning, particularly deep learning, for spectral prediction and the importance of accurately predicting TMT-labeled annotated spectra in proteomics.