Stimulated Raman Scattering (SRS) microscopy is a stain-free, laser-scanning imaging technology that utilizes two coherent laser beams (i.e., the pump and Stokes) to stimulate vibration of chemical bonds in molecules. Different bonds have different resonant vibrational frequencies, and thus SRS microscopy can achieve rapid chemical imaging at high-resolution, enabling live cell imaging and near-instant, stain-free pathological imaging. To increase the ability to resolve different chemical species, multiple Raman wavenumbers can be used with the hyperspectral SRS imaging data. In particular, this approach holds promise for quantifying DNA content, which is important to characterize cancer cell polyploidy. The SRS spectra is the mixture of the spectra of various pure substances present in each pixel so unmixing must be performed to find the relative abundances of these substances. We ran our SRS hyperspectral data of cancer cells through SciPy’s Least Square Error Linear Optimization algorithm (LSQ) [1] but found that it was not able to return the correct DNA content. Our proposed solution to this problem is to use an autoencoder neural network to unmix the spectra. We based the network on the findings in Palsson et al. (2018) [2]. Our initial results show that the network is effective at finding an accurate linear combination, but the noise in the collection of the SRS hyperspectral data significantly increases the number of low error solutions which makes it difficult for the network to find the true linear combination. Future work will be focused on using noise reduction techniques to help the network find the true abundance values.

Stimulated Raman scattering (SRS) microscopy is a stain-free, laser-scanning imaging technology that allows for rapid chemical imaging at high-resolution. When developing cancer diagnosis based on cellular and tissue pathology, it is critical to understand the number, size, and density of the cell nuclei, as well as other metabolic features, such as the lipid droplets. In our research, we compare the U-Net and Mask R-CNN convolutional neural network architectures to segment cell nuclei from SRS images of cultured cancer cells. We also use a modified version of U-Net to identify the centroids of nuclei and lipid droplets. Combining these centroids with a segmentation, we can generate a Voronoi diagram to estimate the size of each nucleus and lipid droplet. Future work will focus on applying these methods to identify and segment various cellular structure in both cells and human cancer tissues with SRS imaging.

Symbolic Regression (SR) is the data driven search for mathematical relations as performed by a computer. In essence, SR is a search over all possible equations to find those which best model the data on hand. Prioritized Grammar Enumeration (PGE) is a recently proposed algorithm which has been shown to have great efficacy and efficiency on the Symbolic Regression problem, using just a single compute core. PGE reformulates the SR problem as a search over a grammar, makes reductions in the magnitude of the search space, and introduces mechanisms for exploring that space efficiently. Notably, PGE provides reliability and reproducibility of results, a key aspect to any system used by scientists at large. In this paper, we enhance the PGE algorithm in several ways. First, we extend PGE to discover differential equations. Second, we incorporate multiple prioritization heaps into PGE, reducing point evaluations while maintaining efficacy. Finally, we decouple the PGE subroutines into a set of services, contain each with Docker, and deploy them onto the cloud. Our algorithm experiments cover a range of dynamical systems from a multitude of domains. and our cloud experiments explore a variety of architectural setups. Our results show PGE to have great promise and efficacy in automating the discovery of equations at the scales needed by tomorrow's scientific data problems.

RDF datasets are an important source of big data. Many of them, however, are too large to fit on a single machine. One approach to address this is to partition the RDF graph across multiple machines, with each component residing on a single machine. A poor partition can incur significant communication costs, however, if as a result many queries involve multiple machines. A number of existing partitioning schemes seek to reduce these costs by finding partitions that avoid cutting edges in the RDF graph. While these can successfully find good partitions the partitioning process itself is often not very scalable, and not capable of handling incrementally-generated RDF data. In this paper, we develop a more scalable, effective and low complexity approach, online graph dataset partitioning, to produce high quality dataset partitions with fewer links between partitions. We show experimentally that it works well in reducing the communication cost of query processing, while at the same time improving scalability of the partitioning itself.