In this article, MOLMAP (molecular maps of atom-level properties) descriptors of reactants, products and reactions were derived from the structures of reactants and products which were represented by physicochemical properties and topological properties of chemical bonds of the compounds. The MOLMAP descriptors above-mentioned were applied to a data set composed of seven classes of photochemical reactions, and three kinds of models were constructed by random forest: (1) The model to predict the type of reaction the reactants produce; (2) the model to predict the type of reaction from which the product can be synthesized; (3) the model to predict the type of the whole reactions. Because the specification of the bonds involved in the reactions was not required in data sets, the MOLMAP descriptors can be used widely. The results obtained in this article were better than those results in our previous researches for the same data set. Therefore, the modification of bond description is helpful for improving the prediction ability of MOLMAP descriptors.

Key words: photochemical reaction classification, MOLMAP AP descriptor, Kohonen SOM, random forest