Abstract

To predict the toxicity of a newly developed drug cheaply, quickly and accurately is one of the top concerns for the drug discovery and development industry. It is believed that even minor improvements will save over US$200 million per new drug. The FDA believes that ‘a new product development toolkit containing … computer-based predictive models is urgently needed’.

Toxicogenomics tries to understand and predict drug toxicity by studying gene expression. However, the state of the art suffers from two very important setbacks (i) it is not model-driven (ii) it doesn’t explicitly help industry decide if a drug should canned or taken further.

Using a human liver cell culture model, SimuGen has demonstrated that with the correct choice of biomarkers, empiric functional data analysis models, and higher level exploratory analysis such as clustering (based on the models), it is possible to predict and describe liver toxicity with greater sensitivity than animal tests. We will work through some of the interesting problems that needed to be addressed to get this right.

SimuGen is a company inspired by the MPhil CompBio course – many of the answers come directly from methods students will become familiar with.