Parallelized analysis of spatial gene expression patterns by database integration
We developed a computational framework for automated integration of a large number of two-dimensional (2D) images with three-dimensional (3D) image datasets located in the standard 3D coordinate. We applied the framework to 2,810 para-sagittal sectioned mouse brain 2D images of in situ hybridization (ISH), archived in the BrainTx database (http://www.cdtdb.neuroinf.jp). We registered the ISH images into the mouse standard coordinate space for MR images, Waxholm space (WHS, https://www.nitrc.org/projects/incfwhsmouse) by linearly transforming them into each of a series of para-sagittal MR image slices, and identifying the best-fit slice by calculating the similarity metric value ({delta}). Transformed 2D images were compared with 3D gene expression image datasets, which were made using a microtomy-based microarray assay system, Transcriptome Tomography, and archived in the ViBrism DB (http://vibrism.neuroinf.jp): the 3D images are located in the WHS.\n\nWe first transformed ISH images of 10 regionally expressed genes and compared them to signals of corresponding 3D expression images in ViBrism DB for evaluating the integration schema: two types of data, produced with different modalities and originally located in different dimensions, were successfully compared after enhancing ISH signals against background noise. Then, for the massive transformation of BrainTx database images, we parallelized our framework, using the IPython cluster package, and implemented it on the PC cluster provided for the Brain Atlasing Hackathon activity hosted by Neuroinformatics Japan Center in Japan. We could identify the best-fit positions for all of the ISH images. All programs were made available through the GitHub repository, at the web site of neuroinformatics/bah2016_registration (https://github.com/neuroinformatics/bah2016_registration).