Abstract

Structures of protein domains offer a dataset to understand the features that mediate their biological function. 3D domain swapping is a structural phenomenon observed in a variety of proteins as a mechanism for dimer or higher oligomeric formation. It is also implicated in neurodegenerative diseases since domain swapping could mediate unlimited oligomerisation. Data on all the available protein structural entries involved in this mechanism have been curated. Information on domain swapping has been obtained from the literature and bioinformatics tools and database resources and reported in a new database, “3Dswap”, for 293 non-redundant protein structures. We propose a quantitative method ‘extent of swapping’ to classify proteins involved in swapping mechanism and observed that large amount of proteins belong to the class of “moderately swapped” proteins. Structural entries from 3DSwap contribute to the ‘positive dataset’ and used for the development of machine learning-based approaches for the identification of sequence and structural features mediating 3D domain swapping. Support Vector Machine (SVM) and Random Forest (RF) based algorithms were developed to predict swapping from structure and mere sequence information. These prediction methods achieve 73% and 64% accuracy, respectively. I will also describe some structural analyses associated with these studies.