Functional Knowledge Transfer for High-accuracy Prediction of Under-studied Biological Processes
Figure 2
Schematic of the functional knowledge transfer.
(A) A functional relationship network is constructed for each organism through Bayesian integration of heterogeneous genomic data (e.g. expression, TF motif binding, physical interaction assays). (B) Functionally analogous gene pairs (i.e. functional analogs) are identified by computing a gene pairwise functional similarity score introduced in Chikina et. al between all sequence homologs. Functional similarity is measured by the statistical significance of the number of common TreeFam gene families in the functional relationship network neighbors of each homologous gene pair. (C) Next, experimentally confirmed biological process annotations for each gene are transferred to its functional analogs. (D) For each biological process the extended set of gene annotations (which include direct gene annotations, if available, and cross-annotated genes) can be used as training examples for machine learning methods (SVM used in this study) to make novel gene membership predictions. (E) Top predicted genes are carried over for experimental validation.