Date of Award


Document Type

Thesis campus only


Computer Science

First Advisor

Matthew A. Hibbs

Second Advisor

Kevin Livingstone


Genetic interactions, especially negative genetic interactions, show the functional relationship between proteins. However, the detection of genetic interactions require dysfunction of genes, which makes it difficult obtain experimentally. Lots of research has been conducted on identifying negative interaction pairs using machine learning methods based on other biological datasets. However, they usually involve using genetic interaction related features, which brings circularity to the method. Therefore, we used physical interactions as our only data source and a support vector machine as classifier to predict negative genetic interactions. We achieved an AUC of 0.906 and accuracy of 0.832 without using any genetic interaction related features, and we have shown that the completion of physical interaction data and genetic interaction data over time has a positive correlation with the model’s performance.