Thursday 25 Jan 2018: Using computer-aided experimental design to understand gene expression dynamics
Dr. Daphne Ezer - Sainsbury Laboratory, University of Cambridge
Harrison 170 14:30-15:30
Human scientists often make ad hoc decisions during the experimental design process, because they cannot interpret the vast amount of data previously collected by other researchers, nor do they have the capacity to optimise all the parameters of their experiments. As a preliminary study, we conducted a survey of 50 biologists in which they were tasked with designing a new experiment, given a set of previous research outcomes. These biologists used a wide range of heuristics for designing their experiments, some of which could bias their research outcomes, suggesting a need for computer-aided approaches in experimental design.
We present two case studies of how computer-aided experimental design might improve research outcomes, both in terms of hypothesis generation and experimental protocol design. Firstly, we describe how machine learning can be used to predict the functional role of regulatory proteins (i.e. transcription factors) that belong to structurally similar families of proteins, and thereby assist in hypothesis generation. This is applied to two of the largest families of regulatory proteins in plants, and we are able to correctly reconstruct gene regulatory networks involved in metal homeostasis and hormone response. Secondly, we provide a computational solution to a common experimental design scenario -- biologists may sample many time points in a preliminary experiment, but then they may only be able to do follow-up experiments on a small selection of time points due to financial constraints. We have developed a tool to help select the most informative time points to sub-sample. This tool is applied to study how light and temperature signals are integrated into the circadian clock in the early morning, saving the lab thousands of pounds and many man-hours of labour. These examples illustrate how data science and optimisation techniques can boost research productivity, thereby enabling biologists to make discoveries more efficiently.