Bioinformatics is the computational arm of life sciences research. It is comprised of computer scientists studying imaging, omics, mathematics, and statistics. In this dissertation I have compiled small pieces from each of these scientific disciplines to develop and deploy two machine learning architectures that address open questions in cancer metastasis and systems biology. In the first experimental chapter we developed a first in class tree-based classifier capable of predicting site specific metastases arising from primary tumors in 16 cancer types. Our model extracts and analyzes the biological determinants of cancer metastases from transcriptomic profiling data to model the biological phenomena of metastatic organotropism. We expanded the core feature selection algorithm into an end-to-end omics preprocessing and feature selection software. We validated our design on the tumor methylation array data by selecting cancer type specific methylation array probes as input features for an artificial neural network. The MetNet architecture was trained to classify cancer types and identify organ of origin in cancers of unknown primary (CUPs). Finally, we developed a high throughput object detection suite for microscopy images of fungal structures. We established and deployed a first in class instance segmentation tool for the identification of arbuscular mycorrhizal fungi colonizing terrestrial plant roots. Our model is available for public use on amazon webservices to support the greater scientific efforts of the fungal biology community.
