There is a large motivation behind finding a better way to develop new drugs and new drug combinations. The current methods of dealing with the ever increasing size of available data in the health industry are insufficient for integrating and using this information for research. In this work I summarize the basics of drug behavior and the properties of drug interactions.
I investigate large databases which integrate several heterogeneous information sources to ease the process of acquiring drug related information and support drug research. Some examples of these databases are presented in this work. The data from these databases can be used to aid patients at home, healthcare professionals and in general increase patient care standard.
Machine learning can be a useful tool for drug research, however it requires the development of appropriate representations of drug molecules and combinations. Utilizing single-molecule representations via molecular fingerprints we can also represent drug combinations with the combination of these and use this combined representation with a machine learning method to construct a model of effective combinations.
In this work I collected data sources that can be used to establish a representation of these molecules, and developed a possible representation method for drug combinations. I calculated these representations and employed them to construct predictive models for discovering new effective combinations. I utilized cross-validation to evaluate these models, then used the ATC classification to search for new combinations with a specific therapeutic effect using one-class prioritization.