Applications in Protein Structure and Function Modeling

Similar to protein-DNA interactions, protein-RNA interactions also perform vital roles in the cell including protein synthesis, viral replication, cellular defense and developmental regulation [145,146]. One major direction in the analysis of protein-RNA interactions is to identify proteins that bind RNA based on features derived from physio-chemical properties of the sequence. A number of published works have focused casting this problem as a binary classification problem using the support vector machines (SVM) classifier to identify proteins that bind RNA [33-35, 51]. Each of these works derived large data sets from the SwissProt database and applied the support vector machines classifier to discriminate protein sequences that bind RNA from all other sequences. Since sequence analysis techniques can identity homologous proteins as having similar function, most of these works reduced the redundancy of the data sets below a certain threshold 40% [35], 25% 

Other studies have focused on identifying surface residues that bind RNA. For example, this problem has been cast in the binary classification setting where the data comprises annotated structures gathered in a fashion similar to DNA-binding residue prediction these works have employed a number of classifiers including neural networks [48,49,52], SVM [53] and Naive Bayes [54]. This problem has also been cast in the structured-prediction setting, which is decomposed into a binary classification problem (solved by neural networks) followed by post-processing 

Protein structure prediction is a central problem in molecular biology and accordingly many techniques have been developed to detect the structural class of a primary sequence. The structure of a protein provides a rich set of features, which can be used to determine the function of the protein. However, the corresponding experimental methods such as x-ray crystallography and nuclear magnetic resonance (NMR) spectroscopy are very time consuming (on the order of months to years) and expensive. Moreover, while there are thousands of protein structures in the PDB, there are still millions of sequences with structures that are yet unsolved. 

There are two main classification systems to organize proteins based on their structure CATH [174] and SCOP [175]. These systems are used to label training data for a number of supervised learning problems found in protein structure prediction. This problem is divided into three subproblems depending on the data 

---