I first developed Deep Learning techniques for several deep learning models. I worked on Bidirectional Recurrent Neural Networks (BRNN), 2-dimensional Recurrent Neural Networks (2D-RNN), N-to-1 Neural Networks (N1NN), and different versions of Convolutional Neural Networks (CNN).
I focused on the models at first and gauged improvements by novel learning techniques.
After this initial assessment, in a second phase I proceeded to test the models on several problems in the protein structural and functional space, including (but not exclusive to) the prediction of protein secondary structure, solvent accessibility, subcellular localisation.
Each of these problems is important, and programs for predicting these properties process hundreds of thousands of queries from all over the world, helping advance research in molecular evolution, the study of protein-protein interaction networks, and computational drug design.