SBIR: Unbiased Behavioral Discovery Platforms

OUSD (R&E) critical technology area(s): Biotechnology

Objective: The DoD seeks to develop unguided behavioral discovery technologies to accelerate detection and medical countermeasure development against current and future threats. These platforms will enable:

1. Quantification of novel behaviors in pre-clinical animal models that are more sensitive than state-of-the-art.
2. High-throughput evaluation of intervention efficacy in animal models with high translational value.

Description: It is impossible to ethically conduct human studies on emerging CBRN threats for the development of effective medical countermeasures (MCM); therefore, the DoD relies on pre-clinical animal models for these efforts [1]. However, animal model behavioral evaluations remain un-modernized, insensitive, and prone to significant variability between operators and protocols. Furthermore, the threat agent doses required to elicit an observable symptomatic response in animal models are much higher than doses that are effective threats to the warfighter. These limitations in reliability and sensitivity hinder rapid emerging threat detection and MCM development. It is therefore paramount to develop more sensitive and unbiased platforms for the discovery and quantification of animal behavior more quickly and at lower levels of effect.

Recent innovations in machine learning have enabled sensitive detection of animal behaviors in indications including pain response, psychiatric conditions and social interactions [2,3,4]. Novel movement patterns and relationships between physical features and movements of animals that are indicative of a response to stimuli can now be quantitatively captured beyond what historical “gold standard” assays can accomplish. The development of technologies that can capture these nuanced novel behaviors and that are agnostic to the stimuli in interest are therefore highly desirable [5]. The end goal of this effort is to provide automated and high-throughput behavioral profiling technologies to accelerate the discovery of both therapeutics and countermeasures that can serve to protect the warfighter.

The evaluation of chemical and biological threat agents poses additional challenges in the sphere of animal behavior: 1) The behavioral feature space sampled should be sufficiently broad to capture unknown phenotypes with unknown kinetics following exposure. 2) The desired speed of response to a given threat agent necessitates high-throughput processing and rapid identification/ detection of features. 3) Use of lethality as an endpoint poses ethical considerations. 4) Platforms should overcome restricted handling and be compatible with high levels of containment. The DoD therefore seeks scalable solutions for the evaluation of toxic endpoints at sublethal doses across animal models spanning insects to mammalian species.

Ultimately, an unbiased behavioral detection system is desired for sensitive evaluation of animal behaviors to support both defense health capabilities as well as chemical and biological defense initiatives throughout the DoD.

Physiologies of interest include but are not limited to:

1. Toxidromes
2. Neuropsychiatric disorders
3. Seizure characterization
4. Traumatic brain injury
5. Neurodegenerative disorders
6. Sleep State

The proposed final solution should:

• Use non-invasive measurements. Surgery is disallowed.
• Not require human annotation of behaviors.
• Identify behavioral responses not previously quantified through human observation for assessing the indication area(s) or interest.
• Demonstrate that incorporating novel behavior detection outperforms relevant state-of-the-art behavioral assays in accurately identifying indication area(s) of interest. State-of-the-art behavioral assays vary by field and should be appropriately included by the proposer.
• Demonstrate increased sensitivity towards identifying exposed animal to subclinical drug dose or identifying acutely diseased animal in a presymptomatic disease state. (Diseases with acute (<2 week) progressions are of interest. Genetic models of disease are not allowed.)
• Integrate multiple behavioral feature spaces: (e.g. movement tracking, orofacial recognition, auditory cues).
• Minimize animal handling and training by the operator.

Phase I

(8 months) The objective of Phase I is to develop a prototype instrument for the automated detection of novel behaviors in one or more animal model (s). To this end, performers should demonstrate that this prototype meets or exceeds the sensitivity and specificity of state-of-the-art behavioral assays for their physiology of interest. Further, phase 1 efforts should demonstrate reproducibility of results on both the same animal cohort and independent cohorts to account for habituation to assay environment and ensure low variability between studies. When considering endpoints that can result from multiple different perturbations (e.g. pain: heat, mechanical, neurological, etc.) demonstrate performance when implemented across multiple stimuli. Success in this phase will also demonstrate dose-sensitivity of the given detection modality (technical justification should be provided for binary classification regimes).

Phase I fixed payable milestones for this program should include:

• Month 1: Report on initial architectures, algorithms, and learning approaches.
• Month 3: Report on acquisition of initial training and test data sets (experimental, simulated or modeled), proposed evaluation metrics, and initial analyses and results.
• Month 5: Interim report describing performance of prototype system.
• Month 8: Demonstrate a prototype capable of automated detection of novel behaviors in one animal model. The demonstration will use naïve animals exposed to performer-selected perturbations of interest.
• Final Phase I Report summarizing approach; prototype architectures and algorithms; data sets; results; comparison with alternative state-of-the-art methodology; quantification of accuracy; quantification of robustness to errors, noise, dropouts, distortions; and quantification of generalizability (ability to model and predict behaviors not explicitly trained on). Report will include drafted instructions/manual for prototype users. 
  
  For performers looking to progress into Phase II, this report will additionally describe the technical feasibility of adapting the prototype in five directions of compatibility:
  • Detecting other physiologies of interest;
  • Use in an aerosolized toxic environment;
  • Use in high-throughput settings;
  • Generation of behavioral barcodes that aren’t reliant on singular behavioral features, and;
  • Use in another animal model.

The report should describe if there will be significant costs associated with any of the compatibility directions. The projected technical readiness level (TRL) shall achieve a TRL of 3 and provide a clear path to Phase II/III and follow-on commercialization.

Direct to Phase II (DP2): Feasibility Criteria

Proposers may apply directly to Phase II if Phase I feasibility has been met outside of the SBIR program. To this end, proposers should provide technical documentation demonstrating automated detection of novel animal behavior for one or more unique physiologies (e.g. pain response, social interaction, neuropsychiatric disorders). Sensitivity and specificity of detected features should match or surpass 1 or more state-of-the-art behavioral assays for a given field. Documentation to substantiate scientific and technical merit/ feasibility may include peer-reviewed journal publications, pre-prints, patents, and/ or proprietary reports. For non-peer reviewed evidence, please provide technical descriptions for experimental design, instrument use, implementation, and requirements such that the data package can be reviewed effectively. Further, DP2 proposers should also include a report including the same information required in a Phase 1 final report:

In this report, provide descriptions of the following: summarized approach, prototype architectures and algorithms, data sets, results, comparison with alternative state-of-the-art methodology, quantification of accuracy, quantification of robustness to errors, noise, dropouts, distortions; and quantification of generalizability (ability to model and predict behaviors not explicitly trained on).

This report will additionally describe the technical feasibility of adapting the prototype in five directions of compatibility:

• Detecting other physiologies of interest;
• Use in an aerosolized toxic environment;
• Use in high-throughput settings;
• Generation of behavioral barcodes that aren’t reliant on singular behavioral features, and;
• Use in another animal model.

Phase II

(24 months) Advance and refine technology developed during Phase I with an optimized design that addresses two or more of the five areas of compatibility outlined in the Phase I report. Fabricate and demonstrate an advanced prototype for laboratory application, verifying that novel behaviors detected by the prototype outperform state of the art for the given indication, animal model, environment, or scale. The 12-month and 24-month demonstration should use naïve animals exposed to a DARPA-selected drug or pathogen which will be chosen based on the performer-selected indications of interest. Provide a report, associated drawings, and control software/source code, if applicable, documenting the theoretical process, design, including any sub-system specifications, performance characterization, projected reliability/maintainability/cost and recommendations to implement the design and implement the system in the target military or commercial application. Deliver a full-scale prototype to support technical testing and evaluation for application in chemical and biodefense response by the end of Phase II. The projected technical readiness level shall be at TRL 6 at the end of Phase II.

Phase II fixed milestones for this program should include:

• Month 2: Report on lessons learned, updated architectures, algorithms, and learning approaches. Report will include summaries of existing training datasets and anticipated additional datasets, quantification of accuracy; and quantification of generalizability (ability to model and predict behaviors not explicitly trained on).
• Month 4: Report on Phase II proposed evaluation metrics, initial analyses and results to include quantitative progress towards addressing at least 1 area of compatibility.
• Month 6: Interim report describing performance of system in 1 area of compatibility.
• Month 9: Demonstrate a prototype capable of automated detection of novel behaviors in 1 area of compatibility. The demonstration will use naïve animals exposed to performer-selected indications of interest. Interim report quantifying system performance, comparing with alternative state-of-the art approaches using machine learning or other conventional methods, and documenting lessons learned.
• Month 12: Demonstrate a prototype capable of automated detection of novel behaviors in at least 1 area of compatibility. The demonstration will use naïve animals exposed to a DARPA-selected indication of interest.
• Month 14: Mid-term Phase II report documenting prototype adaptation to 1 area of compatibility, including architectures and algorithms; methods; results; comparisons with alternative methods; and quantification of accuracy, robustness, generalizability and demonstration outcomes. Report will include drafted instructions/manual for prototype users. Report will summarize planned commercial and government expansion efforts.
• Month 17: Interim report describing performance of system in second area of compatibility.
• Month 21: Demonstrate a prototype capable of automated detection of novel behaviors in at least 2 areas of compatibility. The demonstration will use naïve animals exposed to performer-selected indications of interest. Report quantifying system performance, comparing with alternative state-of-the art approaches using machine learning or other conventional methods, and documenting lessons learned.
• Month 23: Demonstrate a prototype capable of automated detection of novel behaviors in at least 2 areas of compatibility. The demonstration will use naïve animals exposed to a DARPA-selected indication of interest.
• Month 24: Final Phase II Report: Report will document the effort and include information required to transfer the technology to new users, including a complete user manual. The report will document information about the final prototype to include architectures and algorithms; methods; results; comparisons with alternative methods; and quantification of accuracy, robustness, generalizability and demonstration outcomes.

Phase III dual use applications

The proposed technology is intended to dramatically expand the utility of animal behavioral assays in both DoD and commercial markets. Within the DoD, Phase III will be targeted towards chemical and biological defense. Phase III funding should be obtained from either the private sector, a non-SBIR Government source, or both, and will be focused on formalizing/ establishing prototype products for direct transition into chemical threat assessment pipelines. Within the commercial markets, it is expected that technologies developed under this announcement will be directly applicable to the study of disease and drug development. Commercialization efforts for prototypes developed under this announcement will require funding from the private sector, OR a non-SBIR/STTR government source. Work will emphasize advanced development or refinement of technology towards an indication of interest or scaling of technology to accelerate data collection and provide competitive advantage to preclinical development pipelines.

References

1. Office of the Commissioner. (2024, December 16). Animal rule information. U.S. Food And Drug Administration. http://fda.gov/emergency-preparedness-and-response/preparedness-research/animal-rule-information 2.
2. Zhang Z, Roberson DP, Kotoda M, et al. Automated preclinical detection of mechanical pain hypersensitivity and analgesia. Pain. 2022 Dec 1;163(12):2326-2336. doi: 10.1097/j.pain.0000000000002680. Epub 2022 May 11. PMID: 35543646; PMCID: PMC9649838.
3. Mathis, A., Mamidanna, P., Cury, K.M. et al. DeepLabCut: markerless pose estimation of user-defined body parts with deep learning. Nat Neurosci 21, 1281–1289 (2018). https://doi.org/10.1038/s41593-018-0209-y.
4. Pereira, T.D., Tabris, N., Matsliah, A. et al. SLEAP: A deep learning system for multi-animal pose tracking. Nat Methods 19, 486–495 (2022). https://doi.org/10.1038/s41592-022-01426-1.
5. Guo, C.; Chen, Y.; Ma, C.; Hao, S.; Song, J. A Survey on AI-Driven Mouse Behavior Analysis Applications and Solutions. Bioengineering 2024, 11, 1121. https://doi.org/10.3390/bioengineering11111121.

Keywords

Animal Behavior, Biotechnology, Preclinical studies, Animal Models, Medical Countermeasures, Deep Learning, Machine Vision

TPOC-1

DARPA BAA Help Desk

Email

SBIR_BAA@darpa.mil