Statistical and Machine Learning Approaches for Network Analysis

8.10 Applications in Bio-and Cheminformatics

Graph kernels constitute an alternative to vector-based representations of molecules (descriptors) in cheminformatics applications such as ligand-based virtual screening [91] and quantitative structure–property relationships [92]. In bioinformatics, strings as a natural representation of base pair sequences have played a more prominent role from the beginning, but graphs have gained momentum recently, mainly due to advances in systems biology, for example, protein–protein interaction networks.

Although molecular graphs have been used as examples early on, the application of graph kernels in bio-and cheminformatics is still in its beginnings. Until now, a good two dozen retrospective studies (Table 8.2) have been done, covering a moderate range of topics in bio-and cheminformatics (Table 8.3), including the establishment of quantitative structure–activity and structure–property relationships, estimation of absorption, distribution, metabolism, excretion, and toxicity of compounds, and, prediction of protein function. A common theme of most of these studies is that approaches based on graph kernels were either able to achieve state-of-the-art results or surpass them. Unfortunately, the use of different retrospective validation schemes, together with problematic aspects in their design and execution, make an objective comparison of retrospective results in the literature almost impossible.

Table 8.2 Selected References by Study Type.