UMass Machine Learning and Friends Lunch | Main / Registration And Analysis Of In Situ Hybridization Images

Recent advancement in in situ hybridization(ISH) technology has inspired many efforts to collect cellular resolution, genome-wide mapping of mRNA expression. Proper analysis of the resulting images can help us in understanding the role of corresponding genes in the imaged organ at a cellular level. In this paper we present a information theory based method to analyze such images to come up with biologically reasonable hypothesis. Our method consists of registration, expression extraction and expression analysis steps. In registration step, each ISH image is morphed using a a combination of global and local transformation to maximize mutual information with its reference image. Expression values of each anatomical structure is then extracted by applying a semi-automatically created mask corresponding to that structure. In expression analysis step the resulting data is analyzed using a naive over-expression model and bi-clustering to find gene clusters. Resulting registrations suggests that our method is robust over intensity levels and shape variations of the ISH image. Also our functional enrichment results indicate that resulting gene clusters are biologically relevant.