Computational methods for small molecules

Metabolism is the system of chemical reactions sustaining life in the cells of
living organisms. In this thesis we propose four independent computational
studies on metabolic applications. In the first study we propose the first
optimal algorithm for determining atom-to-atom correspondences between
reactants and products in a chemical reaction. We follow up by utilizing the
information of the atom mappings to explore reaction function category
prediction using a novel graph kernel method. The second part of the thesis
discusses methods for analysis of mass spectrometric measurements of
metabolites. The problem of identifying an unknown molecule based on its MS
measurement is of high importance throughout life sciences. We propose novel
kernels and statistical models to this problem.

In this thesis we have applied the latest results from machine learning
research into bioinformatic and chemoinformatic applications.