Without a clear framework for complex and developing labels and context, project progress can be held back.
In the recent past, the idea of implementing a terminology framework for drug discovery projects was commonly dismissed, believed to entail more work than it was worth. But with projects overseen by pharmaceutical and scientific research organisations continuing to be susceptible to data fragmentation, ontology frameworks and semantic technologies are paving the way for interoperability, while helping to boost accuracy and precision.
Research teams are being aided in their AI efforts by combining FAIR data principles, knowledge graphs and organised terminology. As organisations continue to utilise ELNs, LIMS and other informatics systems to progress research and development, ontologies serve as critical infrastructure for scientific data strategies.
A recent online panel discussion organised by Scientific Computing World, featuring experts from session sponsor Revvity Signals along with Rancho Biosciences and Takeda discussed the importance of applied ontologies, and the evolution of mindsets required to transition from siloed legacy data infrastructure towards structured knowledge systems.
Why ontologies are important
Often, research teams made up of various disciplines find themselves communicating the same concepts using different terminology. This language disconnect leaves much to error-prone interpretation that can cause the maintenance of AI models for drug discovery to falter.
But having an ontology in place - a terminology framework that encompasses entire projects - provides a reference point for all staff, regardless of discipline, to stay abreast of drug discovery developments. Standardising the various terms that researchers use for complex ideas is vital towards keeping everyone on the same page.
“At its core, ontologies are really just about creating a consistent way to describe the things we work with, so that both our colleagues - human and machine alike - can interpret them in the same way,” said Samantha Jeschonek, PhD., Product Portfolio Manager at Revvity.
“At its most basic idea, it's a set of structured and standardised terminology that have relationships to each other and rules that govern those relationships.
“In our drug discovery space, that can mean anything from defining exactly what is an assay, to what's an endpoint; how you define a biological system, and then making sure that those definitions are consistent across your discovery pipeline. This allows researchers to break silos and unify their data sets.”
Once relationships between terms are established, they can then be converted into a machine-readable format such as OWL or RDF.
Going from labels to use cases
Merely having names for individual project components is not enough. Research teams need contextual information in order to properly relate developments to each other. AI can then be trained on the data in place in order to help realise new and evolving drug discovery use cases.
“My rule of thumb is, if the ontology is just a controlled vocabulary, and all I'm getting are labels. This is not super helpful to me,” said Julia Fox, PhD., Director of Data & Analytics at Takeda. “This is unless we're talking about the species taxonomy, in which case that is actually helpful because the structure of the hierarchy is meaningful.
“There is a point where you do have to ask, just from your own perspective as a data leader in your environment, ‘What is going to matter for this data at this time in this system?’”.
Dr Jane Lomax, Head of Terminology Services at Rancho Biosciences, added: “There is always an aspect of thinking about use cases that you don't know about yet, though. When I used to build a complex gene ontology, it was often used as flat lists, and we always used to wonder whether we were wasting our time adding all of these logical axioms. But looking back, having this in place is useful for things like reinforcement learning, AI, and explainability.
“While you're there, you may as well encode that information, while you're thinking about it. You may not have a use case right now, but maybe we have one emerging. That can be registered early on before the ontology becomes overly complex. So, thinking about use cases that we don't know about yet is useful.”
The panel collectively shared examples of ready-made ontologies that research teams can start with, and resources if they are new to using them. This included guidance from the Pistoia Alliance; and OBOE, OLS and BioPortal as examples of publicly curated ontologies.
Requirements for bench scientists vs data stewards
The experts were then asked about the differences in needs present when comparing bench scientists that need and use ontologies, with data stewards in research-facing organisations; as well as the importance of both perspectives towards successful adoption.
“You have to approach both perspectives,” said Jeschonek. “I come from the bench, and any time we adopt something new, we have to consider what we get out of it day-to-day.
“For the day-to-day scientist, there has to be some immediate benefit, or it's going to be a real struggle for me to change my ways. Usually, it comes down to asking what kinds of questions you couldn't ask before; whether it was challenging to pull that data together; or what kinds of data sets you wish that you could compare, that you weren't able to do previously.
“It’s all about starting with use cases. That's a great way to lead into ontologies. I find that the biggest convincer is saying, ‘tell me a search you haven’t been able to do’. Now with ontologies, you can do that search. You can pull back all the relevant data, and you have confidence that it's in the right context to be compared.”
When it comes to the perspective of the data stewards overseeing the back end of ontologies, Lomax said: “In some ways, the data stewards and the ontologists, they're sort of servants of the bench scientists, because they're there to organise the data so that the scientists can do their analysis more effectively.
“Everyone's on the same side, but where the tension probably comes in is for those people who have to record your experiments - this is usually the responsibility of the bench scientist. The carrot is doing all of the analysis, but the stick is the admin work required. Modern tooling makes that easier, so really, they shouldn't have to be thinking about ontologies too much. They should just be using a platform that has ontologies embedded into it. You are likely able to use AI to help you choose the right ontology term now without exposing users to OWL files.”
Fox added: “Bench scientists never need to see the ontologies, hopefully. But someone on the back end should understand it well enough to help users guide themselves through the way that data is coming through the system.
“This is often true for clean labs, for example, where you can't bring a notebook in, and we've moved to electronic capture on, say, an embedded iPad. So moving from systems that capture data through manual entry, whether it's typing or loading files or annotating files directly to be loaded, to a more point-and-click approach. This is set up to capture the data you're planning to capture, so you can just go forward with your work.”
By ensuring all researchers involved have the knowledge they need to maintain and utilise the right ontology setup for their projects, the whole team can minimise bottlenecks and stay on the same page.
Research organisations of all sizes can benefit from bridging lexicon gaps across various disciplines, to go from labels to valuable use cases. Ontologies do not need to be made in-house from the start, with various public resources on hand to aid any change management.
Making scientific data meaningful for scientists and AI
The promise of ontologies is not simply that scientists and software can speak the same language, but that scientific knowledge can be connected and understood across disciplines, experiments, and systems.
As AI becomes increasingly central to drug discovery, the challenge is no longer just making data accessible, but making it meaningful. Ontologies provide a foundation for that transition, helping researchers move beyond isolated datasets toward connected scientific knowledge that AI can interpret and scientists can act upon.
The most successful implementations may be the ones scientists barely notice, where meaningful connections emerge naturally within their existing workflows.
More information
For more guidance on how to get the best out of an ontology that fits your workflows, get in touch with an expert at Revvity.
