The main goal of the master's thesis has been to develop prediction models for interactions between RNA and proteins. We have chosen hidden Markov models as our method for modelling and predicting interactions. From our initial data we have extracted representative features and motifs, which we used for building separate models for each experiment. Majority of proteins bind to the same or very similar features and motifs. We have compared the predictive accuracy of models build with two (presence of interaction) and three states (presence and intensity of interaction). Results show that models with two states perform better than models with three states. Merging predictions of multiple single experiment models in combined models, does not improve prediction accuracy. However, combined models perform with high accuracy, and can be used to determine the relations between proteins, such as competition, cooperation and independence with other proteins when interacting with RNA. We have presented hidden Markov models as viable method for predicting interactions between RNA and proteins.
|
