Metabolic Pathway Membership Inference using an Ontology-based Similarity Approach

Predicting metabolic pathway membership with deep neural networks by integrating sequential and ontology information

“…We used multimodal data as input to our model, including the GO terms and k-mer information from protein sequences. While a simplistic approach to represent GO terms is a binary vector with 1 or 0 representing the presence or absence of GO terms in annotation of given gene, our method adopts a scheme from [ 5 ], which considers both of the structure of the GO graph and the information content of the GO terms in building the vector of the gene and their corresponding annotations (Fig. 2 ).…”

“…Various similarity measures have been developed to quantify the semantic similarity of GO terms and applied it in quantitative comparison of functional similarity of gene products, although most of these methods are not developed for metabolic pathway membership inference [ 5 – 10 ]. Essentially, those measures mainly involve two steps of calculation : 1) calculation of GO term similarity, and 2) calculation of protein similarity, based on GO term similarity.…”

“…To infer the protein’s membership in the pathway, the similarity between the proteins are then used [ 7 , 11 ]. More recently, in [ 5 ], a hybrid approach to take into account of both information content of individual GO terms and the whole GO hierarchy with a simple Cosine similarity is shown to be advantageous in both prediction accuracy and running time as compared with other semantic similarity-based methods.…”