Prominent variable detection in lipid nanoparticle experiments : A simulation study on non-parametric change point analysis

Abstract: Images as important sources of information, powered by the newest robotic microscopy technologies, make the volume of data available larger than ever before. From the images, hundreds of variables can be calculated. The medical research group within the HASTE project conducts a LNP (Lipid Nano-particle) experiment that is designed to transport a drug to target cells. In the LNP experiment, microscopy images are taken over time to record if the drug is uptaken. We propose that the non-parametric change point analysis can be used to identify the variable which shows the earliest state change (potentially signifying the drug uptake) among all variables calculated from the images. Two algorithms for non-parametric change point analysis, an agglomerative and a divisive, are studied through simulation leading us to implement the agglomerative algorithm on the LNP experiment data. Furthermore, the simulation results show that the prominent variable detection accuracy improves when more time points are included in the experiment. In the application, correlation is most likely to be detected as the sole prominent variable.