In this research project, the REU students Shabana Khan and Alaesa Hearn have the opportunity to work on exciting and cutting edge research, on drug actions and discovery by using sophisticated machine learning, from abundant genomic data. There are three main components of interest; one is the knowledge available about drug targeting, the second is how to apply neural network based machine learing, and the third has to do with outcome prediction of drugs in the face of variations and uncertainties.

Pre-Mining

In this phase, we use data mining techniques to find "leads", or chemical compounds most likely to work against the disease. Using software from BioSolveIT called FTree, we select batches of promising candidate chemicals.

Current-Mining

After the batches have been selected, the chemicals are analyzed against the molecule that is causing the problem. In the case of cancer or arthritis, this is the protein in the human cell which is malfunctioning. In infections, the drug should bind to any molecule that will interfere with the pathogen's ability to disrupt normal cell functions. The two main considerations are binding and docking. Docking is the correct positioning of the small drug molecule in relation to the much larger protein molecule, rather like a ship pulling into port. Binding refers to the actual interaction of the two chemicals, whether they will create new bonds, and whether the drug will produce the desired effect on the protein.

Post-Mining

Once the candidate molecules have been identified and screened against the protein, other pharmacological considerations must be taken into account. As the drug is distributed into the body, any side effects must be anticipated. The original compound cannot be toxic, nor can any of the secondary products the drug may break down into when metabolized. Further investigations could include any potential for addiction or drug interaction.