FraGNNet: A Deep Probabilistic Model for Mass Spectrum Prediction

The process of identifying a compound from its mass spectrum is a criticalstep in the analysis of complex mixtures. Typical solutions for the massspectrum to compound (MS2C) problem involve matching the unknown spectrumagainst a library of known spectrum-molecule pairs, an approach that is limitedby incomplete library coverage. Compound to mass spectrum (C2MS) models canimprove retrieval rates by augmenting real libraries with predicted spectra.Unfortunately, many existing C2MS models suffer from problems with predictionresolution, scalability, or interpretability. We develop a new probabilisticmethod for C2MS prediction, FraGNNet, that can efficiently and accuratelypredict high-resolution spectra. FraGNNet uses a structured latent space toprovide insight into the underlying processes that define the spectrum. Ourmodel achieves state-of-the-art performance in terms of prediction error, andsurpasses existing C2MS models as a tool for retrieval-based MS2C.