Abstract
The COVID-19 pandemic demonstrated how a novel, elusive, and diffuse biological threat can engender uncertainty and misinformation, and it underscored the need for flexible analytical modalities agnostic to the identity of biological material. Yet even before the pandemic recognition of the limitations of the current, list-based approach, which focuses on known pathogens and biotoxins, and of the importance of agent-agnostic biodetection, was growing within the biosecurity community. In a 2018 report on “Biodefense in the Age of Synthetic Biology,” for example, the National Academy of Sciences stated that “an overreliance on the Select Agent List is a systematic weakness affecting many aspects of the United States” current biodefense mitigation capability." More recently, a group of biodefense researchers proposed the identification and adoption of “bioagent-agnostic signatures (BASs)” as a way of detecting and characterizing not only existing agents but also novel ones, an approach they believe will “enable a more flexible and resilient biodefense posture." Indeed, the future of biodetection requires us to begin developing novel analytics that can identify anomalies and/or characteristics that indicate a potential threat, whether known or unknown, without looking for a specific signature that has been identified previously. To assess potential threats more rapidly, it is critical to develop agnostic artificial intelligence (AI)/machine learning (ML) systems that can be employed for real-time assessment of the nature and source of a perturbation. Such systems should be multiscale and multi-dimensional, integrating sensor data from a range of biological, chemical, and physical application spaces. Emerging deep learning (DL) models demonstrate exceptional promise for identification of discriminatory features within multi-dimensional datasets. DL models have the capacity to recognize and encode highly complex patterns in a wide range of input data modalities, including images, text, and biological/chemical/physical spectra. As such, they can execute a wide range of assessments and determinations that have traditionally required a human operator.
Introduction: The Shift to Agent-Agnostic Detection
Traditional biodetection relies on "knowing what to look for." Current systems like qPCR or standard metagenomic pipelines compare samples against a database of known "Select Agents." This creates a dangerous blind spot for novel, emerging, or synthetically engineered pathogens that are not yet cataloged. Furthermore, standard bioinformatics pipelines involve significant pre-processing—such as base-calling DNA or peak-picking in mass spectrometry—that discards potentially vital information and adds significant latency.
Researchers at Lawrence Livermore National Laboratory (LLNL) advocate for a paradigm shift: moving away from list-based approaches toward Bioagent-Agnostic Signatures (BASs). By applying deep learning (DL) directly to raw instrument data—such as electrical "squiggles" from DNA sequencers or raw mass spectra—threats can be identified based on functional characteristics or anomalies before the data is translated for human readability.
Methodology and Technical Strategies
The PRIMED project is a comprehensive landscape assessment of several AI/ML platforms, utilizing a Machine Learning Technology Readiness Level (MLTRL) framework (on a scale of 0 to 9) to score maturity. Key technical strategies evaluated include:
- Direct Signal Processing: Utilizing Convolutional Neural Networks (CNNs) like SquiggleNet to classify DNA/RNA directly from the raw electrical current (squiggles) of Oxford Nanopore sequencers.
- Protein Embeddings & LLMs: Adapting Large Language Models (LLMs) like BERT and GPT to treat biological sequences as "languages," allowing models to learn the underlying "grammar" of protein function and taxonomic origin.
- Computer Vision for Spectra: Transforming raw Mass Spectrometry (MS) and Raman spectroscopy data into image-like representations and applying transfer learning from models trained on natural images to identify discriminatory features.
- Host-Based Signatures: Analyzing the host's transcriptional response (the human body's reaction to infection) as an indirect, agnostic proxy for the presence of a pathogen.
Key Insights: The Power of Raw Data
- Information Richness: Raw instrument signals contain information that standard processing ignores, such as DNA methylation patterns or signatures of artificial nucleotides.
- Real-Time "Read-Until" Advantage: ML models can classify a DNA strand in less than one second of sequencing. This enables "adaptive sequencing," where hardware can physically eject non-target DNA (like human background) to focus exclusively on potential threats, drastically increasing sensitivity.
- Functional over Taxonomic: Detecting a threat based on what it does (virulence factors) is more robust than what it is (species name), as pathogens often share conserved mechanisms for host infection.
Results and Current Maturity
While the field is promising, most agnostic technologies are currently at an MLTRL of 0.5 to 3.0, indicating they are in early research or proof-of-concept stages. Notable results include:
- SquiggleNet: Demonstrated 75–95% accuracy in distinguishing human vs. bacterial DNA in real-time.
- BERTax: Achieved ~99% accuracy in superkingdom classification, significantly outperforming database-reliant methods on novel sequences.
- SeqScreen: Can assign 32 different "Functions of Sequences of Concern" (FunSoCs) to short DNA reads, providing a functional risk assessment rather than just a taxonomic label.
Limitations and Future Outlook
Several hurdles remain for widespread deployment. Many models are tightly coupled to specific hardware (e.g., Oxford Nanopore), and deep learning models often lack interpretability—the "black box" problem. Additionally, real-world environmental samples (aerosols, wastewater) contain noisy backgrounds that may degrade performance compared to lab-validated samples. Future work must focus on larger, labeled datasets of raw instrument outputs to train more generalizable and interpretable models.
Project Information
Institutional and contact information for the DHS-funded PRIMED project at LLNL. Source, p. 3
Summary updated