Master Thesis




This thesis is focused on deep learning methods for predicting molecular properties from 3D representation, extended by scalar features. The main goal of the thesis will be to experiment with different structures of deep neural network to achieve the highest possible classification accuracy. The neural network will process heterogeneous input in the form of 3D representations of molecules and scalar features. The developed system can be beneficial in future for fast evaluation of new molecules in the process of drug development in pharmacy.

The thesis will be done in collaboration with pharmaceutical company. If the implementation and experiments will be finished within a given deadline, the thesis can be rewarded with 1000, € payment.



The student will work with a framework for molecular properties prediction by deep learning. The code will be extended by the student to experiment with different network architectures and with processing of heterogenous input. The framework is written in python and it is based on Tensorflow and Keras libraries. Labeled data will be provided by the partner company.


  • Knowledge of English language (source code comments and final report should be in English)
  • Knowledge of Python
  • Knowledge of Machine learning algorithms


The project will be implemented in python and the experiments will be executed on a GPU server.


For more information please contact Peter Kán, Hannes Kaufmann.