Machine learning and the quantitative analysis of confocal microscopy with an application to the embryogenesis of drosophila melanogaster

In a multicellular organism, different genes are active in different cells. These patterns of gene activity control the development and differentiation of cells, ultimately deciding the form as an adult organism. Great progress has been made in the study of developmental genetics over the past 60 years. Much of this research is based on data collected at multi-cell resolution. However, it is believed that the details of gene regulation have to be understood at a molecular, rather than cellular, level. Current imaging techniques allow the measurement of multiple gene transcription events at single-molecule resolution, capturing a 3-D volumetric view of the developing organism. Unfortunately, the resulting 3-D images are too complex to allow for manual analysis. In other words, the data is available, but it is difficult to extract and to translate into a usable form. We develop a system which transforms images of multiple gene activity patterns into discrete models which capture the different combinations of nascent transcription occurring within individual cells. Using simple computer vision algorithms and sophisticated machine learning, we have developed image segmentation and counting algorithms that are adaptive. We employ active learning to engage experimentalists to teach the system their preferences, and we use these preferences to tune the parameters of these simple computer vision algorithms. This system, coupled with the experimental methods of the McGinnis and Bier Labs, is used to transform a large image archive of early stage Drosophila embryos imaged using multiplex Fluorescent in situ Hybridization (FISH) into a quantified model of Drosophila transcriptional repression at the detail of a single-molecule. With this model we study the regulation of the transcriptional state of individual nuclei within regions of developing Drosophila embryos, directly measuring the impact of Snail repressor on transcription of genes short gastrulation (sog), ventral nervous system defective (vnd), and rhomboid (rho).