A Fast Neural Network for Isotopic Charge State Assignment

Electrospray ionization (ESI) mass spectrometry is an essential technique for chemical analysis in a range of fields. In ESI, analytes can produce multiple charge states, which must be correctly assigned for identification. Existing approaches to charge state assignment can suffer from limited accuracy and/or poor speed. Here, we developed a fast deep learning neural network to perform isotopic cluster charge assignment. The performance of our algorithm, IsoDec, was demonstrated on top-down proteomics spectra collected on diverse instruments. On these highly complex individual spectra, we found that IsoDec produces similar sequence coverage to existing software tools but with improved accuracy. Importantly, this performance enhancement stems directly from the neural network charge assignment approach, not simply improved scoring and filtering of isotopic clusters. Finally, when applied to large top-down proteomics data sets, we discovered that IsoDec produces proteoform-spectrum matches with a better combination of coverage and accuracy. Overall, IsoDec provides a compelling demonstration of the potential of lightweight neural networks in mass spectrometry data analysis for diverse applications.