Choosing the right model in cell-based screening

At first glance, cell-based screening can look like a question of technology. Which assay format should you use? Which platform is newest? Which model is most advanced?

That was not the main lesson from ELRIG’s recent Cell-based Screening 101 webinar. Instead, the session returned again and again to a more practical question: what is the right model for the question you are trying to answer?

For early career professionals, that is an important distinction. Cell-based screening is often presented as a ladder of progress, with simple two-dimensional models at one end and highly complex systems such as organoids and organ-on-chip platforms at the other end. It is tempting to assume that the more complex the model, the better the science. The discussion in this webinar painted a more grounded picture.

The speakers described cell-based assays as sitting in an important middle ground within drug discovery. They offer more physiological relevance than assays built around purified proteins alone, while remaining more tractable and scalable than animal studies. That balance helps explain why cell-based assays are used across so many stages of discovery, from target validation through to hit identification, lead optimisation and preclinical decision-making.

A recurring message was that simpler models still matter.

Two-dimensional cell systems and immortalised cell lines continue to play a central role, particularly in high-throughput settings where teams may need to screen very large numbers of compounds. In that environment, speed, reproducibility and cost are not secondary concerns. They shape whether an assay is genuinely useful.

One of the clearest principles from the session came from the discussion of high-throughput screening: assays should be as simple as possible, but as complex as necessary.

That may sound like a neat phrase for a slide deck, but it captures a real discipline in assay design. Early in discovery, researchers may simply need to know whether any chemistry can modulate a target in a robust and measurable way. At that stage, a simpler cellular system may be entirely appropriate. Later in the process, the question changes. Teams may want to understand whether a compound has the desired effect in a more disease-relevant setting, whether it behaves differently in a three-dimensional structure, or whether it can distinguish between healthy and diseased biology. That is where more advanced models start to earn their place.

The webinar brought this to life through several examples. A fibrosis assay showed how even a relatively simple two-dimensional model can still generate meaningful functional readouts. By measuring collagen deposition and markers of myofibroblast activation, researchers can assess whether a treatment is changing behaviour associated with fibrotic disease. The point was not that this simple model fully recreates the disease state. It does not. The point was that it can still answer a useful question well.

Another example came from cell painting, a phenotypic imaging approach that looks not at one target or one protein, but at broader changes in cell morphology. By staining major cellular structures and analysing patterns across many features, researchers can begin to generate phenotypic fingerprints of compound effects. This opens up a different kind of readout, one that is more comparative and less narrowly targeted. It also hints at the direction the field is moving in, with richer datasets and more multidimensional ways of understanding biology.

That shift raises an obvious question: if better and more complex models are now available, why not use them all the time?

The panel’s answer was refreshingly honest. Advanced systems come with real trade-offs. Primary cells can be more biologically relevant, but they are harder to scale and more variable. Induced pluripotent stem cell-derived models can open doors to cell types that are otherwise difficult to study, but they are costly and technically demanding. Three-dimensional systems such as spheroids and organoids may capture aspects of tissue architecture and disease behaviour that are absent in flat cultures, yet they are slower, harder to image, harder to standardise and less suited to
very large screens.

That does not make them less valuable. It simply means they have to be used with intent.

The session also touched on something that tends to get less attention in public conversations about drug discovery: robustness. A more sophisticated model is not automatically a better model if its variability makes the results difficult to interpret. This becomes especially important when patient-derived material is involved. If an assay is not robust enough, it becomes hard to tell whether a difference in response reflects real biological heterogeneity or just inconsistency in the system itself. That is not a small technical detail. It goes to the heart of whether a model can support confident decisions.

The conversation then moved into the wider future of the field. Artificial intelligence and predictive screening were discussed with interest, but also with caution. The panel did not present AI as magic dust sprinkled over a difficult workflow. Instead, they framed it as a tool whose value depends on the quality and quantity of underlying data. For well-characterised targets, predictive approaches may already help reduce the number of compounds that need to be tested experimentally. For newer and less understood targets, the need for high-quality biological data remains substantial.

The same measured tone carried through the discussion on non-animal methods and regulation. There was clear optimism about growing momentum behind advanced in vitro systems and stronger regulatory interest in data generated from these approaches. But again, the emphasis was on evidence, reproducibility and confidence.

Progress will not come from novelty alone.

It will come from demonstrating that these models help teams make better decisions and generate data that others trust. For anyone early in their career, that may be the most useful takeaway from the webinar. Cell-based screening is not just about tools, platforms or fashionable model systems. It is about judgment. It is about understanding the biological question, recognising the practical constraints, and choosing an assay that is fit for purpose.

That may be less glamorous than claiming every future lies in organoids, AI or fully automated biology. But it is far more useful. And in drug discovery, useful usually wins.