Secondary structure motif determination in ncRNA via graph kernel based computational models

ncRNA which is a functional molecule but yet not translated into protein has significantly taken importance in the field of bioinformatics, therapeutics chemoinformatics and for the advancement of science. The nucleotide composition and its structure (identity of paired and unpaired nucleotides) determine the function of ncRNA. Key analytical tools such as folding, alignment and clustering algorithms rely on energetic considerations to generate the accurate response to specific queries as they are designed. In reality, these algorithms become inaccurate while considering the non-linear effects with underlying assumptions (energy additivity), when violated. To overcome these eventualities, one can formulate key parameters in terms of nonlinear functional dependencies that can be learned from known examples (or parts of examples) or from suboptimal RNA structure prediction. Given the importance of the structural element in ncRNA these methods should ideally be able to work in structured domains i.e. they should be able to accept input graph data structures. The methods will belong to the family of kernel machines, since this class of algorithms allows to use heterogeneous features and to accept complex instances such as sequences of graphs as input. The aim of the thesis is to develop computation model capable of identifying subgraphs within the ncRNA folding graph that are characteristic of biological functions. Further subject them to kernel models to improve the RNA secondary structure and its prediction in terms of accuracy.