Seminar "Deep learning applications in proteomics mass spectrometry"

Thursday, 13^th of October at 18.00 there will be a seminar on the topic "Deep learning applications in proteomics mass spectrometry".

The speaker will be William Stafford Noble, Professor, Department of Genome Sciences Department of Computer Science and Engineering University of Washington.

Abstract: Tandem mass spectrometry analysis of biological samples yields large, complex data that is ripe for analysis using machine learning techniques. In this talk, I will describe two recent projects in which we used mass spectrometry data to train deep neural network models. The first project involves training a Siamese network to project peptide mass spectra into a learned latent space in such a way that spectra generated by the same peptide are close together and vice versa. We used the trained model, called GLEAMS, to perform large-scale spectrum clustering, and we used the resulting clusters to explore the dark proteome of repeatedly observed yet consistently unidentified mass spectra. The second project involves training a transformer model to translate a mass spectrum, represented as an ordered series of peaks, into an amino acid sequence. The resulting de novo peptide sequencing tool, called Casanovo, substantially outperforms existing methods for this important problem.