Automated Identification and Assessment of a Brain Microbiome
Abstract
Automated Identification and Assessment of a Brain Microbiome While the brain has long been thought of as a sterile environment, preliminary manual analysis of electron microscopy (EM) imagery of mammalian cortex has suggested the presence of bacteria in both healthy as well as diseased brain tissue. Early results from germ-free controls were reported which suggest this was not simply due to sample contamination. This controversial finding conflicts with the widely-held neuroscientific understanding that the blood-brain barrier, a cellular boundary which isolates the central nervous system, is impermeable to bacteria in a healthy individual. If these early findings are confirmed, they may radically alter our understanding of disease in the brain. The major barrier preventing large-scale analysis is the potentially immense scale of EM imagery (a single cubic millimeter of tissue can reach petabyte-scale), the specialized human expertise required in order to manually scan these datasets for bacteria (prior research in this area has employed teams of expert bacteriologists), and the expense of collecting new datasets. This proposal aims to overcome these issues by employing a machine learning approach wherein a computer vision algorithm is trained to identify bacteria in public-domain images of brain tissue. A training dataset will be initially seeded by human annotation and augmented with EM images of bacteria collected from non-neural tissue. The output of the algorithm will then be used to refine and retrain the machine learning approach using input from human proofreaders. Finally, the machine learning system will be deployed at scale on cloud computing resources to search for bacteria in public-domain datasets. The scientific objective of this project is to test the hypothesis that bacteria are more widespread in nervous system tissue than previously believed, while providing a computational tool to detect and quantify bacteria in brain tissue at scale. We will build on existing work at JHU/APL on storage and processing of large scale EM datasets, as well as electron microscopy of bacterial samples. Public domain datasets are available in online repositories, and can be processed using cloud computing resources. We will retrain machine learning approaches commonly used to find subcellular structures such as synapses and cell membranes for bacteria detection. The discovery of a brain microbiome would fundamentally change our understanding of brain disease, but there is also the possibility the preliminary observations are due to experimental error or contamination. To answer this critical question, this work will outpace manual human annotation. Success with this project will aid in confirming or denying the detection of bacteria in large volumes of neuropil. If confirmed, analysis of a neural microbiome could provide fundamentally new approaches to studying neurodegenerative diseases such as AlzheimerÕs disease as well as mood disorders such as depression. Approved for public release; distribution is unlimited
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 09, 2020
- Source ID
- W911NF2010166
Entities
People
- Jordan Matelsky
Organizations
- Army Contracting Command
- Johns Hopkins University
- United States Army