Predicting tissue-specific protein functions using multi-part tensor decomposition

摘要: Abstract Proteins are complex molecules that play many critical functions in the human body. They expressed different tissues body where their vary depending on tissue they in. The disorder of protein interactome affects biological which results diseases. Therefore, understanding and assessing tissue-specific is essential for disease diagnostics therapeutics. However, it a hard task as requires laboratory experimentations resources expensive have limited scalability. Thus, multiple computational approaches were developed to predict functions. These managed provide predictions with high scalability efficiency. still suffer from rates false positives. In this work, we propose new method predicting using tensor factorisation multi-part embeddings. We model proteins, functions, corresponding tensor, apply learn scores all possible protein-function associations each studied tissues. then show by experimental evaluation our outperforms state-of-the-art models margin 33.3% 13% area under precision recall ROC curves respectively.