Skip to main content
ARS Home » Research » Research Project #442872

Research Project: Biologically-inspired Odor Olfactory Detection Guided Machine Learning (BIODOG-ML): Odor Detection from Canines to Instrumentation

Location: Research Programs

Project Number: 3022-32000-018-015-S
Project Type: Non-Assistance Cooperative Agreement

Start Date: Sep 15, 2022
End Date: Sep 15, 2024

Biological threat surveillance for pathogens of national security is limited by the accessibility and flexibility of current biodetection systems. Biological threats that can impact the agricultural industry and create significant economic losses as well as threats to public health represent a category of complex and inherently variable targets for detection systems. The first line of defense in the detection of biological targets necessitates a rapid mobile technology to direct support resources, such as law enforcement, security teams, public health professionals and laboratories, toward suspect areas, materials and/or individuals. Programs, such as those outlined in the recent 2021 National Blueprint for Biodefense, do not readily have the capability to detect biological agents in real-time and state “More than five years after we released A National Blueprint for Biodefense [2015], the United States remains at catastrophic biological risk,” indicating a critical security gap. As the handling of any infected materials or cultures pose great risk these materials are relatively inaccessible for use in developing canine training capabilities. The Cooperator has a long history of training canines for detection of various chemical and biological homeland security threats. The FBI has invested significant funding into the development of a canine training aid, the Polymer Odor Capture and Release (POCR) mobile training aid, specifically to be used to train canines to detect compounds that are hazardous and restricted in nature which are secured in controlled environments, such as high biosafety containment facilities, and thus the canines cannot have direct access to them for training. Canine detection of plant pathogens has long been used for disease surveillance and control, however minimal work has been done in animal pathogens. Further very little work has been done to attempt to identify which compounds or pattern of compounds form the odor ‘fingerprint’ that the dogs may be alerting to. The objective of this agreement is to leverage previous federal and academic investment toward establishing a model for BSL3+ virus detection using a repository of previously collected samples from a variety of livestock species (cattle, swine and sheep) infected with bovine viral diarrhea virus as a prototype. This agreement will combine Auburn’s canine expertise and POCR mobile training aid with ARS’s expertise in virology, immunology, volatile organic compound analysis, artificial intelligence and machine learning as a proof-of-concept study to determine the ability to detect volatile biomarkers of infection using proton mass spectrometry. This is a unique approach to develop biologically inspired systems of informed machine learning. Using the dog as an intelligent biological sensory system to inform the establishment of classifiers should improve the instrument optimization and achieved sensitivity. This multi-disciplinary approach may provide a non-invasive, high-throughput and highly sensitive bio detection system.

The approach would leverage prior federally sponsored controlled animal study results and sample repository of a BSL2 pathogen. In a previous study for viral detection using trained detection canines, Cooperator's Canine Performance Sciences demonstrated a detection capability for bovine viral diarrhea virus (BVDV) in live culture and successful discrimination of that virus from similar viruses in live culture with detection dogs. This model virus represents an agent that affects multiple species and has closely related viruses of foreign animal disease significance. BVDV is in the Pestivirus genus alongside classical swine fever virus and border disease virus and belongs to the Flaviviridae family encompassing viruses such as Japanese Encephalitis virus, yellow fever virus, Zika virus, Dengue virus and West Nile virus. The Cooperator has a field research unit for BVDV (BSL 2) herd-based projects, which enables the field operations testing of canines under real-world conditions. This unit is unique in its access and capabilities to support this work. Use of this selected viral model for developing training protocols using a restricted and hazardous agent set of conditions (handling and access mimicking in BSL 3+) provides a robust means for a proof of concept across a multi-species platform and under operationally relevant conditions (field-side live animal testing). The training and testing of canines as biosensor using a polymer-based training aid system (POCR) will start with the foundational learning and performance in a new cohort of detection canines. Initial canine training will be on the POCR-system (initial task learning) on a non-relevant training surrogate. Once proficient at being trained on the task the detection canines will progress to training on BVDV-POCR system as a target virus through culture-POCR follow restricted (BSL-3+) protocols and be required to demonstrate successful discrimination of the target (e.g. virus) from other non-targets (e.g. other viruses) in POCR system. Following proficiency in training, detection canines will be evaluated on performance in generalization testing to the live target from other non-targets. Proficiency in finding live materials will advance the canine cohort onto evaluation of performance in generalization for operational sampling (e.g. nasal discharge, saliva, tears, etc.) and in field testing (e.g. farm-side testing and environmental sampling, etc.). The final evaluation of performance will be to test ability of trained canines to the target across species (in-vivo and in-vitro samples), across infective windows and define any associated limitations. The use of detection canines to code samples based on response metrics will support further use of canine data and materials to concurrently evaluate volatile odor signatures through proton mass spectrometry incorporating AI machine learning. The evaluation of samples used as training materials can establish potential unique VOC viral ‘fingerprints’ and build an odor signature library which can be applied toward advancing this methodology toward a non-invasive odor sampling for biodetection and surveillance for targets of concern.