Pretrained spectrum representations for diverse tandem mass spectrometry proteomics tasks

Mass spectrometry is the dominant technology in the field of proteomics, enabling high-throughput analysis of the protein content of complex biological samples. Due to the complexity of the instrumentation and resulting data, sophisticated computational methods are required for the processing and interpretation of acquired mass spectra. Machine learning has shown great promise to improve the analysis of mass spectrometry data, with numerous purpose-built methods for improving specific steps in the data acquisition and analysis pipeline reaching widespread adoption. Each of these tasks rely on a shared fundamental understanding of the information present in a mass spectrum. Here, we propose a transfer learning framework, CasanovoTL, which utilizes pretrained spectrum representations, derived from a de novo sequencing model trained on a massive proteomics dataset, rather than training a spectrum encoder from scratch for each task. We show that using these pre-trained spectrum representations significantly improves our performance on the four downstream tasks of spectrum quality prediction, chimericity prediction, phosphorylation prediction, and glycosylation status prediction. Finally, we perform end-to-end finetuning, as well as multi-task training, and find that finetuned spectrum representations further improve performance on each individual task. Overall, our work demonstrates that a pretrained spectrum encoder for tandem mass spectrometry proteomics trained on de novo sequencing learns generalizable representations of spectra, improves performance on downstream tasks where training data is limited, and can ultimately enhance data acquisition and analysis in proteomics experiments.