by Karin Tjarnlund, VP, Protein Research, Discovery & Medical, Cytiva
Drug discovery is changing quickly. Research teams are working with more complex targets and more diverse therapeutic modalities and are subject to increasing pressure to make better decisions earlier to reduce the risk of failure later in the clinic. At the same time, the emergence of artificial intelligence (AI) is reshaping how scientists generate and prioritize ideas from disease target identification to de novo binder design.
AI is not replacing the experimental work of drug discovery. Instead, it is changing where that work creates the most value. Rather than relying on broad, blind, yes-or-no threshold-based assays to identify promising candidates, teams can increasingly use computational approaches to prioritize which hypotheses deserve deeper experimental characterization.
That shift does not reduce the need for high-quality data. It raises the bar for it. As more hypotheses are generated computationally, researchers need information-rich experimental evidence to validate which candidates truly engage the intended target and warrant further study. That is why molecular interaction analysis is becoming more important, not less.
Modern discovery spans many molecular formats, from antibodies and multi-specifics to antibody drug conjugates, membrane protein targets, fragments, small molecules, and targeted protein degraders. Across these areas, the central question is not simply whether two molecules interact. Rather, it is how they interact, how strongly and selectively they bind, how stable the interaction is, and whether that interaction profile supports the intended biological or therapeutic objective.
Surface Plasmon Resonance (SPR) has helped answer these questions for many years. What is changing now is not the fundamental value of the technique, but the context in which the data is used: discovery teams need interaction insights earlier, at greater scale, and in forms that can be interpreted, compared, connected, and applied quickly across programs and decision points.
Why interaction profiles matter
Interaction data is important across multiple stages of the drug discovery workflow. First, it can help confirm that compounds bind to the intended target, whether they come from a traditional binary screen or an AI-selected hit list. This validation step helps reduce false positives and gives teams greater confidence in which candidates deserve further study. From there, information-rich interaction data can reveal how validated hits engage the target: how strong the interaction is, how quickly it forms and dissociates, how specific it is, and whether the resulting profile supports further characterization and lead optimization.
Different therapeutic approaches create different analytical questions, and each requires more than a simple yes-or-no answer.
In fragment-based discovery, weak binding can be valuable when it provides a rational starting point for optimization. A small molecule may show promising affinity yet lack the residence time or kinetic profile needed for the intended effect. Two biologic candidates may appear similar in an initial assessment but behave differently when characterized in greater depth. In biosimilar development, interaction analysis can help assess whether a proposed biosimilar and originator product engage relevant targets or receptors in a similar way.
Having this information earlier can help teams focus resources on candidates with stronger interaction profiles and remove problematic molecules before they create downstream risk. That matters because failure later in development is often linked to insufficient target engagement, weak specificity (or off-target effects), or interaction profiles that do not support the therapeutic product profile.
In practical terms, this shifts interaction analysis from a single confirmation step to a source of evidence that can guide decisions across hit validation, candidate prioritization, and lead optimization, helping teams decide not only whether a molecule binds, but whether its interaction profile supports continued investment.
AI changes the demand for experimental evidence
AI-enabled discovery can help teams generate and prioritize more hypotheses, but those hypotheses only become useful when they are tested against reliable experimental evidence.
A model may suggest which molecules to test, which variants to optimize, or which candidates to deprioritize. The decision to advance a molecule, however, still depends on experimental evidence that the interaction is specific, reproducible, and relevant in the intended biological context.
In that sense, AI helps reduce the number of low-probability candidates entering the lab, but the candidates that remain still need to be validated and understood in greater depth. Interaction data can help confirm whether AI-selected hits truly engage the intended target, reduce false positives, and reveal whether the binding profile supports further optimization. By helping teams deprioritize poorly behaved compounds earlier, interaction data can create room for additional rounds of lead optimization, support more parallel programs, and help de-risk molecules before they advance toward the clinic.
As AI becomes more embedded in discovery, reliable interaction data becomes part of the feedback loop between computation and experimentation. The stronger that feedback loop, the better positioned teams are to move from model-generated possibilities to experimentally grounded decisions.
From faster experiments to faster decisions
The challenge is generating information-rich interaction data early enough, and at the scale required, to influence discovery decisions. AI-assisted design, expanding molecular formats, and tighter timelines are increasing the number of hypotheses that need to be validated and prioritized.
Higher-throughput and more parallelized systems can generate more data, faster, helping teams validate more hits, compare interaction profiles, and prioritize promising molecules earlier. But capacity alone is not enough.
As workflows become faster and generate increasing amounts of data, the bottleneck shifts from data generation to data analysis, interpretation, and usability. More measurements are valuable only if researchers can use them efficiently, compare results across candidates and studies, and move the evidence into the systems where decisions are made. The broader point is not faster data for its own sake; it is faster decisions from integrated, information-rich workflows.
This is where data quality and data structure become as important as experimental output. Interaction data loses value if it is difficult to retrieve, compare, or reuse across studies, systems, and teams. For modern workflows, the issue is not simply having more data, but having data that is reliable, contextualized, and structured enough to support future learning cycles.
Looking further ahead, more automated and AI-orchestrated lab-in-a-loop workflows will drive further efficiencies and compression of the drug discovery workflow. But realizing the potential of an autonomous, lab-in-a-loop future depends on robust instruments, reliable consumables supply, consistent assay performance, and software environments that connect experimental execution with data interpretation. Automation only creates value when the underlying systems are dependable and robust enough to support it.
The next phase of molecular interaction analysis
Drug discovery will continue to rely on both computational insight and experimental evidence. AI can help teams generate more ideas and prioritize them more intelligently, but experimental characterization remains essential for testing those ideas in the real world. The future will depend not only on better models or faster instruments, but on the ability to connect experimental evidence, data infrastructure, and workflow reliability into systems that support confident decisions.
Organizations that benefit most will be those that treat interaction data not as isolated assay output, but as connected evidence that moves through integrated workflows and informs decisions across the discovery lifecycle.