Link to bioRxiv paper: http://biorxiv.org/cgi/content/short/2020.09.13.295105v1?rss=1 Authors: Tarn, C., Zeng, W.-F., Fei, Z.-C., He, S.-M. Abstract: Spectrum prediction using deep learning has attracted a lot of attention in recent years, and has been used in the analysis of data-dependent and data-independent acquisition mass spectrometry. Although existing deep learning methods have greatly increased the prediction accuracy, there is still considerable space for improvement, which is presently limited by the difference of instrument types or settings. In this work, we use the few-shot learning method to fit the data online to make up for the shortcoming. We evaluate the method using 9 commonly used datasets, where the instruments include Velos, QE, Lumos, and ABSciex, with NCEs being differently set. Experimental results show that, on nearly all of the datasets, pDeep3 achieves significantly higher prediction accuracy, with extermly low costs. For example, on a trypsin dataset Pandey-Velos of Orbitrap Velos from Human Proteome Map, with only 100 randomly selected peptide-spectrum matches, it increases the accuracy from 49% to 85%, and only takes 7 seconds using CPU. Further, we also show that our design is sufficiently effective to fill the accuracy gap, and not demanding on data quality. Finally, using OpenSWATH and EncyclopeDIA, we present the potential advantages of pDeep3 for DIA data analysis. The source code of pDeep3 is available at http://pfind.ict.ac.cn/software/pDeep3. Copy rights belong to original authors. Visit the link for more info