Logo image
Back
The complete interactome with 5-D GPU accelerated analysis, visualization, and UI for biological microscopy applications
Dissertation   Open access

The complete interactome with 5-D GPU accelerated analysis, visualization, and UI for biological microscopy applications

Eric C. Wait
Doctor of Philosophy (Ph.D.), Drexel University
Jun 2019
DOI:
https://doi.org/10.17918/gv6v-vm87
pdf
Wait_Eric_201977.91 MBDownloadView
PDF Open Access Open Access (License Unspecified)

Abstract

Graphics processing units Visualization--Data processing Bioinformatics Computer Engineering
Microscopes now have the ability to collect highly complex data. This thesis presents new algorithms and software packages for analyzing, visualizing, and interacting with biological microscopy image datasets. These new software packages provide functionality not available with existing methods. Along with new functionality, the methods described here are of higher accuracy compared to existing methods. These new software packages provide significant improvements in computational efficiency and usability. Contributions described within this thesis are biologically motivated with examples given throughout. The analysis presented has enabled new biological discoveries along with new mathematical analytic techniques. This thesis is broken into three distinct areas: analysis of interactions between organelles describing the complete interactome, a hardware accelerated image processing library, and software packages providing visualization, validation and analysis. First the complete interactome describes a new method of quantifying the interactions of all of the organelles within a single cell. Where prior methods focus on pairwise interactions, methods described here include a robust quantification method for these high order interactions. Secondly, a hardware accelerated image processing library and framework called Hydra Image Processor (HIP). HIP is capable of swiftly processing arbitrarily large datasets. The HIP framework automatically (without user intervention) distributes data and schedules processing across all available graphic processing units. In addition, HIP processes the edges of images with higher precision than currently available methods. Lastly, visualization, validation, and analysis tools are presented. Visualization of high dimensional datasets captured by modern microscopes is non-trivial. Integrating processing, segmentation, and tracking results embedded is also a challenge. Software packages presented here perform all three of these task within the scripting environment of MATLAB for fast prototyping and swift user interface generation. Validation tools are also described that incorporate user input to inform unsupervised algorithms and reduce underlying segmentation and tracking results significantly. All methods and contributions of this thesis are biologically motivated. Within the three areas described above are biological examples where these methods contributed significantly. They range from describing the high-order interactions within a single cell, to lineaging stem cells in 5D, to the reconstruction of large montage images of the Subventricular Zone. The adoption of these methods will enable future biological findings within highly complex datasets captured by modern and future microscopes.

Metrics

28 File views/ downloads
20 Record Views

Details

Logo image