pathway2vec

pathway2vec is a software package advancing the reconstruction of metabolic pathways from genomic sequence information, which is crucial for understanding the regulatory and functional potential of cells across various levels of biological organization. Traditional methods for metabolic pathway reconstruction have been primarily gene-centric, relying on mapping annotated proteins to known pathways using reference databases. However, pathway-centric approaches infer pathway presence through heuristics or machine learning and offer a novel avenue for hypothesis generation within biological systems. These approaches, however, often depend on complex rule sets or feature information that may not be readily available or known.

pathway2vec employs representational learning to automate feature generation for pathway inference. The software constructs a three-layered network comprising compounds, enzymes, and pathways. This structure facilitates intra-layer interactions and captures significant betweenness interactions across layers. By leveraging this layered architecture, pathway2vec efficiently learns a neural embedding-based low-dimensional representation of metabolic features, encapsulating relevant relationships essential for pathway analysis.