A method for constructing protein semantic networks using MEDLINE abstracts is proposed. The publications retrieved by the context search for protein names (relevant) and the related publications were used. The proposed method is based on estimation of semantic connectivity between proteins. The connectivity score was calculated as a function of the number of relevant or related papers found for a pair of proteins. This score was used to construct a semantic network for 150 human proteins belonging to five different metabolic pathways. Analysis of the network demonstrated that the proteins involved in associated molecular processes formed the subgraphs with a high edge density.