Bringing functional single-cell analysis into drug discovery
Sarah Taylor, VP of Science at Lightcast, discusses how functional single-cell analysis can complement existing technologies, improve early-stage drug discovery decisions and help researchers identify promising candidates by measuring what cells do.
DDW: You spent nearly nine years at 10x Genomics taking genomic technologies from concept to launch, and you have just joined Lightcast as VP of Science. What did you see in functional single-cell analysis that persuaded you to make the move now?
ST: My experience at 10x showed me what happens when you give researchers a technology that’s accessible and scalable: the field moves faster than anyone predicts. Single-cell sequencing changed how we understand cells, and more recently spatial technologies have added context to that understanding. But even a billion-cell atlas or a high-resolution spatial map built on omic data still relies on inferring function. We can learn in extraordinary detail what every cell is and how it’s configured, but what it does remains a prediction. Functional measurement felt like the next layer, the technology to do it at single-cell resolution is now ready, and Lightcast is building it in a way that gives researchers direct access to it.
DDW: The field has become very good at measuring what cells are. Measuring what they do is a different proposition. Where does that distinction actually matter in practice, and where does it not?
ST: The distinction matters wherever a decision hinges on behaviour: does this cell kill its target, secrete what we need, respond to this molecule, and if so, how strongly? In antibody discovery, cell therapy and mechanism-of-action studies, identity is a good guide but not the complete answer. It matters less where identity genuinely is the question, such as cataloguing tissues, finding new cell types or mapping lineages. Omics is superb at that. The point isn’t that omics falls short, it’s that some decisions need a measurement, not a prediction.
DDW: Bulk assays average across populations, and rare cells carry much of the biology worth finding. Can you provide an example of where population-level data concealed something that single-cell resolution would have revealed?
ST: There are many examples in the literature of cell types, states and niches that went unrecognised until tissues were analysed with technologies that provide single-cell resolution. The discovery of the pulmonary ionocyte, a rare airway cell type accounting for most CFTR expression, is a landmark example (Plasschaert et al., Nature 2018; Montoro et al., Nature 2018), and more recently a study of the heart revealed several distinct cellular niches (Kanemaru et al., Nature 2023). In my own work in colorectal cancer, we found a clonally expanded T cell population making up 11% of one tumour’s T cells, sitting near tumour cells in chemokine-rich niches (Oliveira et al., Nature Genetics 2025).
Bulk assays would have masked all of these findings.
DDW: Single-cell sequencing and spatial biology have transformed how researchers map tissues and cell states. Where does functional analysis sit alongside those methods? Is it complementary, or does it answer a question those platforms were never built to answer?
ST: Complementary. Functional analysis answers questions the other platforms were never designed to answer. Single-cell and spatial profiling technologies tell you what cells are present, what state they’re in and where they sit, based on the snapshot in time in which they were frozen or fixed. Functional analysis is performed on live cells and tells you which cells are interacting, how those cells behave, and what they secrete or bind to when challenged.
DDW: Researchers often treat transcript abundance as a stand-in for cellular behaviour. What does single-cell sequencing capture accurately in making that leap, and where does the inference fail?
ST: Transcript levels often do point in the right direction. But it’s well known that mRNA abundance is a step removed from protein levels, secretion, signalling and timing, and it says nothing about how cells interact. The inference breaks down most visibly in antibody discovery, where a B cell may carry a promising sequence, but whether the antibody it produces binds, neutralises or triggers the right response is a functional question. Likewise, in cell line development, expression profiles don’t reliably tell you which clones will actually perform. Progressing the wrong cell or candidate based on a proxy is a costly mistake.
DDW: Functional readouts have been technically possible for years, yet they remain a specialist activity rather than a routine one. What has held the field back? Instrumentation, workflow complexity, data interpretation, or something about how programmes are designed?
ST: It’s been a combination of factors. The instruments have historically been expensive and hard to use, which kept them in specialist hands. The workflows are complex and don’t slot easily into existing programmes, and interpreting the data has often required expertise most labs don’t have in-house. On top of that, the companies building these tools have tended to focus on big pharma, so there’s been less published evidence, fewer independent users and less community knowledge to draw on. Technologies become routine when a wide base of researchers can use them, publish with them and build on each other’s work. That’s the gap we want to help close.
DDW: Late-stage failure is one of the biggest costs in drug discovery, and the biological limitations behind it often only emerge once a programme is far advanced. If functional data arrived earlier, which decisions would teams make differently, and how far upstream does the effect reach?
ST: Most candidates that enter clinical development never reach patients, the majority of these failures come down to a lack of efficacy or unmanageable toxicity, and I believe many trace back to biology that wasn’t understood early.
Earlier functional data changes several decisions. In target validation, you confirm that engaging the target produces the intended cellular response, not just that it’s expressed. In hit selection, you rank candidates by what they do rather than only how they bind. In mechanism of action, you see how different cells respond. And at candidate progression, you’re less likely to carry forward molecules with functional liabilities that would only emerge late. The effect reaches furthest upstream at hit selection and early triage, because that’s where better decisions are cheapest and late surprises most expensive.
DDW: Envisia uses optical electrowetting-on-dielectric (oEWOD) to control individual droplets deterministically, which means researchers can hold a cell, combine it with other cells or reagents, observe its behaviour, and recover it. Why does that combination matter scientifically, and what does it let a researcher ask that other formats do not?
ST: The capabilities create the most value when combined. Control over how cells and reagents meet makes the experiment designed rather than merely observed: the researcher decides what a cell encounters, in what order and when, instead of watching whatever happens across a population. Observing the interaction in a single droplet ties a specific response to a specific cell and the stimulus that triggered it. Recovering that cell means the cells of interest can then be expanded, cloned or sequenced. That lets a researcher ask which individual cells respond to a defined challenge and then isolate them for further work.
Other formats typically force a trade-off: you measure the response but lose the cell, or you keep the cell but can’t control the interaction. Closing that gap is what’s new.
DDW: What evidence does the field need to see before functional single-cell analysis becomes a standard step rather than an optional one?
ST: A technology becomes standard when skipping it starts to feel like a risk. We saw this with single-cell sequencing. It didn’t replace bulk overnight. It became routine once researchers saw it reveal biology that bulk had averaged away, and once the workflows, tools and published examples made it practical to use. Functional measurement needs three kinds of evidence to make the same move. First, that it changes decisions: concrete examples where functional data led a team to progress a different candidate or drop one earlier. Second, that the result is trustworthy: reproducibility, and validation against methods the field already trusts. Third, that it fits the way programmes already run: results that integrate with existing omic data and computational models, and show that adding functional layers improves what those models can predict. Working with customers lets us test all three against real questions, not idealised ones. And showing that function strengthens the analyses people already do is a far better route to standard practice than asking anyone to replace them.
DDW: Look ahead five years. If functional measurement becomes routine, what does a discovery workflow look like, and what will researchers stop doing?
ST: Functional measurement will sit near the start of discovery workflows alongside omics, so teams choose cells, targets and candidates based on measured behaviour from the outset. These measurements will also change how AI is used in discovery. Models trained on omic data predict function, and they are only as good as the ground truth they learn from. Large-scale functional data, with each measured response linked to the identity of the cell that produced it, gives those models something to learn from and be tested against. Prediction, experiment and feedback then become a loop, with each round of functional measurement improving the next round of predictions.
What becomes obsolete is the habit of treating a proxy as an answer, for example progressing a long list of candidates on binding or expression alone and sorting them out later and relying on bulk assays to rank things that really depend on rare or heterogeneous responses. I’d also expect fewer low-information screening rounds, because earlier, richer data should let teams narrow to better candidates sooner.
