a) Deep Learning based mapping from molecule drawing of chemical structures to chemical properties (mined from ChEMBL). Design, fit and validation of a chemical properties mapper from Marvin-based molecule drawings of chemical structures. We propose the use of a deep autoencoder to build a representation of molecule drawings through direct drawing or Marvin JS definitions. Posterior mapping will be design, fit, and validated through machine learning or a Deep Neural Network. Data from CheMBL.
b) Complex Workflows Through Natural Language Processing. This project will include, specification design for a system for automated complex data analysis through voice commanding under Amazon Alexa. The project will use an improve an initial framework for the implementation of general complex workflows through natural language processing developed at the B2SLab.
Both projects will provide a tuition for the student. Selection criteria will be based from academic CV and personal interview.
Please send request to alexandre perera at upc dot edu, with subject B2SLab Master’s thesis projects offering 2017.