Date of Award


Degree Type


Degree Name



Computer Science

First Advisor

Rueda, L.


Biological sciences, Applied sciences, Cath, Domain, Machine learning, Obligate and non-obligate, Pfam, Protein-protein interaction




Prediction and analysis of protein-protein interactions (PPIs) is an important problem in life science research because of the fundamental roles of PPIs in many biological processes in living cells. One of the important problems surrounding PPIs is the identification and prediction of different types of complexes, which are characterized by properties such as type and numbers of proteins that interact, stability of the proteins, and also duration of the interactions. This thesis focuses on studying the temporal and stability aspects of the PPIs mostly using structural data. We have addressed the problem of predicting obligate and non-obligate protein complexes, as well as those aspects related to transient versus permanent because of the importance of non-obligate and transient complexes as therapeutic targets for drug discovery and development. We have presented a computational model to predict-protein interaction types using our proposed physicochemical features of desolvation and electrostatic energies and also structural and sequence domain-based features. To achieve a comprehensive comparison and demonstrate the strength of our proposed features to predict PPI types, we have also computed a wide range of previously used properties for prediction including physical features of interface area, chemical features of hydrophobicity and amino acid composition, physicochemical features of solvent-accessible surface area (SASA) and atomic contact vectors (ACV). After extracting the main features of the complexes, a variety of machine learning approaches have been used to predict PPI types. The prediction is performed via several state-of-the-art classification techniques, including linear dimensionality reduction (LDR), support vector machine (SVM), naive Bayes (NB) and k-nearest neighbor (k-NN). Moreover, several feature selection algorithms including gain ratio (GR), information gain (IG), chi-square (Chi2) and minimum redundancy maximum relevance (mRMR) are applied on the available datasets to obtain more discriminative and relevant properties to distinguish between these two types of complexes Our computational results on different datasets confirm that using our proposed physicochemical features of desolvation and electrostatic energies lead to significant improvements on prediction performance. Moreover, using structural and sequence domains of CATH and Pfam and doing biological analysis help us to achieve a better insight on obligate and non-obligate complexes and their interactions.