Prediction of molecular substructures from mass spectograms
Pieter-Jan Drouillon, Hendrik Blockeel · Lirias · 2007
This paper describes work in progress. We propose a possible approach to predicting the structure of a molecule from its mass spectrogram. The main idea is to cluster molecules based on their mass spectrogram. On these clusters one can perform a frequent subgraph mining algorithm to find the most frequent substructures in these molecules. Substructures that are much more frequent in one cluster than in others are likely to have an important influence on the mass spectrogram. Once these substructures have been identified, they can be used in a second clustering step to improve the clustering, after which a new search for frequent substructures in the new clusters can be performed. This can be repeated until the process stabilizes, which should lead to clusters that are coherent with respect to mass spectrograms as well as those molecular substructures most related to them. We discuss two variants of the approach, both of which remain to be validated empirically.