Laboratories are entering a new phase of digital maturity, where tools for automation, informatics, and AI are no longer just capturing data but actively shaping how it is used. As these systems become more connected and intelligent, expectations around data have shifted. It is no longer enough to store results—data must be structured, contextualized, and ready to move across workflows, systems, and teams. This evolution places new emphasis on usability, interoperability, and the ability to turn data into a reliable operational asset.
In this Q&A, Nicolas Triballeau, PhD, and Zev Wisotsky, PhD, of Revvity Signals discuss where data workflows tend to break down, what FAIR principles look like in practice, and how lab leaders can strengthen metadata, systems, and standards without adding unnecessary burden to their teams.
Many labs generate large volumes of data but still struggle to use it effectively. Where do you see the biggest breakdown between data collection and data usability?
The breakdown almost always occurs at the point of context—or rather, the absence of it. Labs are generally good at generating data, but the data is often captured without the metadata that gives it meaning. A file of tabular results, for example, tells you very little if it doesn't describe what was measured, under what conditions, using which protocol, or by whom.
Compounding this is the prevalence of rigid data schemas—predefined structures that force scientists to fit rich, multifaceted experimental results into limited input fields. When data is bound too early into a fixed structure, it loses the nuance and flexibility needed to answer questions that haven't been asked yet. Scientific discovery is inherently unpredictable; data captured today may need to inform analyses that weren't imaginable when the experiment was run.
The result is data that is abundant but effectively inaccessible—siloed, poorly modeled, and stripped of the scientific context that would make it genuinely useful for analysis, collaboration, or AI.
When you think about FAIR principles in a lab setting, what does that actually look like at the workflow level?
FAIR principles—Findable, Accessible, Interoperable, and Reusable—are often discussed at the policy or infrastructure level, but their real value is realized at the bench, in the daily workflow of the scientist.
At the workflow level, FAIR means that when a researcher records an experiment, they are prompted—ideally in a frictionless way—to capture the metadata that contextualizes their results: the method used, the conditions, the parameters, and the protocol version. It means that data carries unique, persistent identifiers so it can be located and retrieved reliably. It also means data is structured in a way that is both human- and machine-readable, and that it shares enough common structure to be compared across experiments, teams, and even organizations.
Critically, FAIR at the workflow level also means interoperability across disciplines and geographies. This is where ontologies become essential.
An ontology captures a common understanding of a given domain of knowledge, including entities, terminologies (controlled vocabularies), various types of relations between entities, and describes them in a structured format that is readable by both humans and machines. For example, a typical cell ontology lists different primary cells, cell lines, cell types, their organ of origin, their species, and sometimes their genetic modifications. Scientists in different countries or different fields may use different terminology to describe the same cell line—what one lab calls a " Human Embryonic Kidney 293" and another may use “Graham 293”.
Without ontological mapping to reconcile these differences, statistical comparisons across datasets become unreliable, and the "Reusable" dimension of FAIR breaks down entirely. Ontology support allows scientists to compare results that use different vocabulary to describe the same underlying concept, making cross-study analysis genuinely meaningful.
Lab Quality Management Certificate
The Lab Quality Management certificate is more than training—it’s a professional advantage.
Gain critical skills and IACET-approved CEUs that make a measurable difference.
What role do LIMS, ELNs, and instrument software play in enabling or limiting FAIR data practices?
These systems are the primary point of contact between the scientist and the data, which means they are either the greatest enablers or the greatest barriers to FAIR compliance.
An electronic laboratory notebook (ELN), for instance, has the opportunity to prompt scientists to capture metadata at the moment it is most accurate and most top of mind: during or immediately after the experiment. When an ELN is configured with customizable, requirable experiment properties, it creates a structured incentive for scientists to document context early and consistently—without adding significant burden to their workflow.
Conversely, systems that are rigid, siloed, or that treat data capture as an afterthought actively undermine FAIR principles. Instrument software that exports raw data without contextual metadata, or LIMS that enforce fixed schemas with no flexibility for evolving research needs, create exactly the kind of data quality problems that make AI and advanced analytics difficult to apply.
The most effective systems are those that integrate data capture, schema management, and metadata enrichment into a single, connected workflow—reducing friction for the scientist while ensuring the data is structured, annotated, and ready for downstream analysis.
Interested in lab leadership?
How can labs improve metadata quality without adding significant burden to staff?
The key is to make metadata capture a natural part of the scientific workflow rather than an administrative task. Several practical principles can help achieve this:
- Leverage ontologies to standardize terminology. Ontological frameworks enable labs to align different terms—from various disciplines, geographies, or conventions—with shared, standardized concepts. This greatly improves the consistency and comparability of metadata across studies without requiring scientists to manually reconcile terminology differences.
- Use configurable, required fields. Instead of relying on individual discretion, data management systems can enforce key fields before a new entity is registered or an experiment record is closed. Most fields should use drop-down menus based on standardized terminology, ensuring consistency with minimal manual effort. While free-text inputs allow flexibility, they should be limited since they often introduce noise and irregularities.
- Register assays with rich metadata. In drug discovery and similar fields, experiments are typically runs of well-optimized assays with fixed conditions that ensure reproducibility and throughput. Capturing metadata once at the assay level saves time for bench scientists and prevents downstream errors across thousands of experimental runs.
- Capture metadata at the point of work with semi-automated tools. Not all information is defined in advance—such as timestamps, locations, or plate IDs—but systems can automatically fill many of these details. Metadata recorded during the experiment, when context is clear and details are fresh, is far more reliable than retrospective entries.
- Adopt flexible schema design. A “late-binding schema” approach lets data structures evolve closer to the time of analysis instead of being fixed at capture. This reduces the need to predict future use cases and allows teams to correct modeling errors without rework—saving time while maintaining data integrity.
For a lab that wants to improve how it uses its data, what are the first practical steps they should take?
Start with an honest audit of your current data quality. Ask: Can we find our data reliably? Does it carry enough context to be understood by someone who wasn't in the room when the experiment was run? Can it be compared meaningfully with data from other studies or other teams?
From there, a few foundational steps tend to have the highest impact:
- Define a metadata strategy. Identify the minimum set of contextual information that must accompany every experimental result—methods, conditions, parameters, identifiers—and build that into your data capture workflows.
- Invest in ontology support. Particularly for organizations working across multiple teams, therapeutic areas, or geographies, ontological mapping is not optional—it is the infrastructure that makes data truly findable and interoperable. Without it, data that appears comparable may be describing subtly different things.
- Move away from rigid schemas. If your current systems force data into fixed structures that can't evolve, consider approaches that allow for late binding of schema—where data structure can be adjusted as research questions change, without losing the underlying data or its context.
- Integrate your tools. Data that flows seamlessly between an ELN, a data management system, and analytical tools retains its context at every step. Fragmented systems create fragmented data.
- Make it easy for scientists to do the right thing. The best data management practices are the ones scientists will actually follow. Systems that reduce friction, prompt good behavior, and integrate into existing workflows will always outperform those that rely on manual compliance.
How does FAIR-aligned data enable better decision-making, automation, or AI initiatives?
AI is only as powerful as the data it consumes. This is perhaps the most important and most underappreciated truth in the current enthusiasm around AI in life sciences. Organizations that attempt to apply AI and machine learning to poorly structured, lacking context and data consistency, find that the results are disappointing—or worse, misleading.
FAIR-aligned data directly addresses this. When data is findable and accessible, it can be efficiently retrieved and aggregated for analysis. When it is interoperable, it can be combined across studies, teams, and systems without manual reconciliation. When it is reusable, it can serve as training data for ML models— and the quality of that training data directly determines the quality of the model's predictions.
Metadata enrichment is particularly critical for AI. Tagging data with rich contextual information allows ML models to be trained on specific, relevant subsets of data rather than on generalized datasets that dilute the signal. For example, having detailed metadata on compound toxicity allows a model to make more precise predictions and steer researchers away from unproductive design spaces.
Ontology plays a specific and important role here as well. For AI to draw valid comparisons across datasets—whether for statistical analysis, pattern recognition, or predictive modeling—it must be able to recognize that two differently labeled data points are describing the same underlying concept. Ontological mapping provides that semantic bridge, enabling AI to work across the full breadth of an organization's data rather than being limited to datasets that happen to use identical terminology.
The downstream benefits extend to automation and decision-making as well. Well-structured, contextualized data enables automated workflows, reduces the time scientists spend searching for and interpreting data, and supports more confident, evidence-based decisions at every stage of the research process.
