AI-Ready Flow Cytometry Meets NIST’s AI Principles

As laboratories explore artificial intelligence-ready systems, bottlenecks in data flow and interpretation are giving way to automated, intelligent experimentation

Drug development is entering the era of AI-native biology, where autonomous experimental design, multimodal single-cell analytics, and generative AI workflows depend fundamentally on the quality, reproducibility, and standardization of upstream wet-lab steps. The National Institute of Standards and Technology (NIST) has been spearheading flow cytometry standardization and has been a leading voice in shaping how AI is measured, evaluated, and responsibly adopted across domains. In 2025, NIST’s engagement with the AI community highlighted several key lessons that align well with how laboratory automation platforms should approach AI integration. AI cannot outperform the biological noise created by inconsistent sample preparation. This positions algorithmic laboratory operations, particularly the Curiox Pluto Code, a software-defined representation of wash physics and mixing dynamics, as enabling technology for AI-ready data pipelines. It does so by transforming sample preparation from a manual, error-prone process into a deterministic, platform-agnostic, algorithmic operation, precisely aligned with NIST expectations for measurement assurance, reproducibility, and cross-site comparability. As flow cytometry becomes a centerpiece of multimodal AI models, the convergence of automation, data standards, and AI will define the next decade of research and development.

AI Is Limited by Wet-Lab Variability

NIST has emphasized that AI systems must be rigorously evaluated not just for raw performance, but for reliability, robustness, and validity across real-world laboratory contexts, such as immunophenotyping panel design for pharmacodynamic endpoints. This means moving beyond static metrics to lifecycle evaluation practices that capture how AI behaves during repeated use and under variable conditions. Despite breakthroughs in machine learning, i.e. automated gating, generative cytometry, multimodal integration, the flow cytometry field faces a structural barrier: wet-lab variance remains the dominant source of error, overshadowing algorithmic limitations. 

Key findings echoed across the 2025 NIST workshop include:

• Sample preparation (pipetting, washing, mixing) causes orders of magnitude more variability than cytometer hardware.
• Metadata and FCS file inconsistencies limit interoperability and inhibit cross-study modeling.
• Lack of standardized reference datasets constrains AI generalizability, including foundation models.

Peer-reviewed literature corroborates these conclusions:

• Batch effects propagate through ML pipelines even after normalization (Maecker et al., 2010; 2012).
• Cytometry variance arises largely from staining/wash differences, not instrumentation (Konecny et al., 2024).
• Centrifugation induces cell loss and apoptosis, distorting population frequencies (Brown et al., Sci Rep., 2018).

New drug development programs increasingly demand true multi-site reproducibility, AI-ready, FAIR-compliant datasets, and platform-independent experimental behavior. To achieve this, automation that behaves as code should be adopted. This shift is critical for accelerating discoveries, but adoption hinges on technologies that seamlessly integrate instruments, data, and analytics into algorithmic workflows. 

Workshops and ongoing dialogues have underscored the need for shared vocabularies, standardized evaluation frameworks, and consistent metadata practices. These help ensure that AI-driven insights are comparable, interpretable, and reproducible, critical attributes for drug discovery pipelines where decisions must withstand regulatory scrutiny. NIST’s risk management work encourages organizations to map, measure, manage, and govern AI systems to mitigate unintended consequences and align technology deployment with ethical and operational goals.

Why Algorithmic Wash Physics Matter: The Scientific Foundation of Pluto Code

At its core, the Curiox Pluto Code is a standardized integration framework for flow cytometry automation, enabling data to move effortlessly from acquisition to clean, normalized formats that can feed directly into machine learning and AI models. This not only improves throughput and reproducibility but also lays the groundwork for real-time, adaptive control of experiments based on AI insights. Pluto Code represents a new class of bioprocess algorithms: not scripts for automation robots, but computational models of fluid dynamics, mixing kinetics, and reagent interactions that ensure reproducibility regardless of platform. 

What are the primary sources of flow cytometry variance? Based on NIST analyses:

• wash and mixing steps are the largest contributors to noise,
• operator-driven variance disables automated gating,
• incomplete antibody removal compromises marker fidelity,
• centrifugation causes population distortion.

By encoding wash physics in an algorithm, it:

• ensures consistent shear conditions,
• eliminates pellet re-suspension artifacts,
• standardizes kinetic profiles across instruments,
• creates a stable upstream environment for AI.

Why Sample Workflow Standardizations Are Critical for AI

Traditional centrifugation disproportionately damages fragile immune subsets, introduces unmeasured mechanical variability, and produces non-Gaussian distributions detrimental to ML training. AI systems mistake these artifacts for biological signals. Removing centrifugation is not a convenience; it is a precondition for trustworthy AI inference in immunology, oncology, and cell therapy. By marrying standardized flow cytometry workflows with AI systems evaluated through NIST-aligned criteria, laboratories can create fully traceable, transparent, and trustworthy AI data pipelines. This accelerates early drug discovery stages, like high-content screening or immune profiling and supports decision-making in translational and clinical phases. In the coming years, combining automated cytometry, AI-ready data infrastructure, and rigorous evaluation frameworks will help drug developers uncover deeper biological insights and reduce experimental variability.

Bench + Bytes is a column written by Charles River Scientist Christoph Eberle, PhD, and Melvin Lye, Senior Director, Scientific Affairs and Product at Curiox Biosystems. It is hosted by Eureka, Charles River's scientific blog.