Drug discovery is entering an era defined by unprecedented resolution in understanding cellular behavior. Traditional approaches often focus on one or a few biomarkers, giving a narrow view of how a compound affects cells. But recent advances in high-content imaging and computational biology are enabling a more holistic perspective — capturing the complexity of cellular responses in a single experiment.

One technique at the forefront of this shift is Cell Painting, a morphological profiling assay that visualizes multiple cellular structures simultaneously using fluorescent dyes. By generating rich, high-dimensional data on organelle organization, shape, and texture, Cell Painting moves beyond hypothesis-driven screening, offering an unbiased window into cellular phenotypes. This allows researchers not only to identify potential drug effects and toxicity earlier but also to uncover unexpected mechanisms of action and opportunities for drug repurposing.

To explore how Cell Painting is reshaping the landscape of drug discovery, DDN spoke with Jonathan Sexton, Associate Professor of Internal Medicine and Medicinal Chemistry at the University of Michigan, about the science behind the method, its practical applications in the lab, and the challenges and opportunities it presents for translating cellular insights into safer, more effective therapeutics.

What is Cell Painting and why is it important for drug discovery?

Cell Painting is a morphological profiling assay that uses a multiplexed set of fluorescent dyes to visualize specific cellular compartments — nucleus, nucleoli, endoplasmic reticulum (ER), mitochondria, Golgi apparatus, cytoskeleton, and plasma membrane — simultaneously. What makes it fundamentally different from traditional high-content screening is its unbiased, hypothesis-free nature. Rather than designing an assay around a specific target or biomarker, we let the biology speak for itself.

In drug discovery, this shift in philosophy is transformative. Instead of asking "does this drug inhibit kinase X?" we ask "what is the comprehensive cellular response to this compound?" This approach is particularly powerful for drug repurposing, one of my core research focuses, because compounds with unexpected activities in one disease context often reveal therapeutic potential in another. The morphological fingerprint can identify novel mechanisms of action, flag early toxicity signals, and cluster structurally dissimilar compounds that share biological effects, all from a single assay.

How does Cell Painting differentiate from traditional phenotypic assays in terms of data richness and predictive power?

Traditional phenotypic assays typically measure a handful of endpoints — viability, a reporter signal, perhaps a biomarker intensity or protein translocation event. They answer narrow questions but can miss the broader cellular context. Cell Painting extracts thousands of morphological features per cell: size, shape, texture, intensity distributions, spatial relationships between organelles, etc. A single well can yield millions of measurements.

This dimensionality fundamentally changes what's predictable. In my work on drug-induced liver injury using patient-derived liver organoids, for example, we can detect organelle-level perturbations — mitochondrial fragmentation, ER stress patterns, phospholipidosis signatures — long before they manifest as overt toxicity. Compounds that look identical in a viability assay may have completely distinct Cell Painting profiles, revealing different mechanisms and different risk profiles. The technique effectively converts microscopy images into high-dimensional, computable biological data that machine learning (ML) models can leverage for predictions we could not make by visual inspection alone.

How is Cell Painting used in your lab?

In my lab, Cell Painting serves as a primary screening modality, but we've adapted it in ways that extend well beyond standard implementations.

Mechanism of action discovery: We compare novel compound fingerprints against annotated reference libraries — including the JUMP-CP consortium data — to generate mechanistic hypotheses. This is particularly valuable for repurposing candidates where the original indication tells us little about potential new uses.

Patient-derived organoid platforms: We've moved beyond immortalized cell lines to apply Cell Painting to intestinal and hepatic organoids derived from patient tissue. This is technically demanding — 3D structures require different staining penetration strategies and more sophisticated segmentation — but it bridges the gap between throughput and physiological relevance.

AI/ML pipeline development: We use the rich Cell Painting datasets to train ML models that can detect subtle phenotypic changes invisible to the human eye. In our work on drug-induced liver injury (DILI), we are developing classifiers that predict clinical hepatotoxicity by analyzing morphological signatures in patient-derived liver cells. These models accurately identify drugs with known DILI risk. For example, using our approach, we were able to detect significant liver toxicity in several drugs that recently failed clinical trials, including inarigivir for chronic hepatitis B. The models revealed that its toxicity was driven by mitochondrial damage, highlighting how ML predictions can uncover specific mechanisms of harm.

Barrier function prediction: One of our current projects uses machine vision to predict functional outcomes in intestinal organoids for inflammatory bowel disease (IBD). We have developed models that can accurately detect leaky gut, a condition where the intestinal barrier is compromised, allowing bacterial antigens to cross into the body and trigger inflammation — a key driver of IBD. By linking Cell Painting morphological features with measurements of barrier integrity (transepithelial electrical resistance), we created a predictive model that can guide drug discovery for IBD by identifying compounds that improve intestinal barrier function.

What are some of the challenges or limitations of working with Cell Painting today?

Several practical challenges persist despite the technique's maturation:

Data infrastructure: A single large screen generates terabytes of images and hundreds of gigabytes of tabular data. We run our analysis using CellProfiler pipelines that output parquet files for downstream processing, but storage, versioning, and compute resources remain significant operational challenges.

Batch effects: Cell Painting is very sensitive to experimental variation — staining duration, reagent lots, cell passage, even plate position effects. These batch effects can obscure true biological signals. We've invested heavily in normalization strategies and plate design to mitigate this, but it requires constant vigilance.

Biological interpretation: We can extract a feature called "radial distribution of ER intensity at 50 percent radius" and detect how they change in response to a compound, but interpreting what these changes mean biologically remains challenging. The field is improving at linking specific feature patterns to known cellular processes, but much work remains. Combining Cell Painting with high-throughput transcriptomics can help reveal the underlying mechanisms, but it takes significant expertise to translate these data into a clear, interpretable molecular mechanism of action.

Translating to complex models: Applying standardized Cell Painting protocols to organoids introduces new variables: penetration depth, optical sectioning requirements, segmentation of irregular 3D structures. The protocols that work beautifully in 2D monolayers require substantial optimization for our 3D patient-derived organoid systems.

Cell Painting as a technique is around a decade old now. How has this tool improved over that time?

The evolution has been remarkable, driven by improvements in high-content imaging hardware, biology and chemistry tools, and advancements in computational approaches for analysis.

Imaging throughput: We use modern high-content imaging systems, specifically the Yokogawa Cell Voyager 8000 (CV8000), which allow rapid collection of high-quality images. The CV8000 uses water immersion objective lenses that greatly improve spatial resolution and signal-to-background ratios, and its four scientific CMOS (complementary metal-oxide-semiconductor) cameras capture four channels at the same time. Older Nipkow confocal spinning disc systems, by contrast, suffered from a drastic loss of signal that made them unusable for dyes/stains that weren’t very bright. The Yokogawa dual microlens spinning disc focuses light more efficiently through the pinholes, producing stronger signals even from weaker dyes. This lets us image faster, achieve higher resolution, and maintain robust signal intensity without compromise.

3D model compatibility: The adaptation to organoids and spheroids represents perhaps the most important biological advance. Early Cell Painting was validated exclusively in 2D immortalized lines. We're now successfully applying it to patient-derived 3D cultures, which introduces challenges in staining penetration and confocal imaging requirements but gains physiological relevance that dramatically improves translational value.

Reagent improvements: Dye formulations are now more stable, and innovations like the Mitsui gas-permeable InnoCell plates — which we are testing for organoid experiments — help maintain high-quality 3D structures across large-scale screens.

Analysis infrastructure: Nuclear and cellular segmentation has improved dramatically thanks to neural networks such as the CellPose 4 SAM model, which can achieve nearly perfect segmentation automatically, without any user input.

Additionally, the JUMP-CP consortium has established reference datasets and standardized analysis workflows that didn’t exist five years ago, greatly improving our ability to compare data across instruments and batches. At the same time, CellProfiler has matured, and the field has developed widely accepted approaches for normalization and quality control.

How do you see Cell Painting integrating with omics technologies such as transcriptomics, proteomics, or metabolomics to create a more holistic view of cellular states?

Cell Painting provides something the omics technologies cannot: direct visualization of the phenotypic outcome of a perturbation, whether that be a small molecule drug or a genetic perturbation like CRISPR interference. Transcriptomics captures what the cell plans to do, proteomics informs what machinery the cell has available, and metabolomics indicates what reactions are occurring. Cell Painting shows the phenotypic result of a perturbation that anchors the analysis of other omics technologies with visually detectable changes that improve our ability to decipher the mechanism of a perturbagen.

This integration creates powerful validation loops. When we see mitochondrial fragmentation in Cell Painting, we can ask whether the metabolomics show impaired respiration and whether the transcriptomics show stress response pathway activation. Concordance across modalities dramatically increases confidence; discordance points to interesting biology worth investigating.

For our DILI research, we're pursuing exactly this kind of integration — correlating morphological signatures in liver organoids with transcriptomics and clinical phenotypes. The goal is triangulation: no single data type is definitive, but convergent evidence across modalities builds a case for a mechanism that any single approach would miss.

One practical advantage of Cell Painting is that it is non-destructive: the same cells can later be harvested for molecular profiling, allowing true paired measurements from identical samples.

From your perspective, what are the key steps needed to standardize Cell Painting across the pharmaceutical industry so that it becomes a routine part of early drug discovery pipelines?

Industry-wide adoption requires infrastructure, not just enthusiasm.

Reference datasets: The JUMP-CP consortium has made enormous progress here, providing annotated compound libraries with standardized profiles. These serve as ground truth for benchmarking and enable labs to calibrate their own data against community standards.

Protocol harmonization: We need validated standard operating procedures that specify dye concentrations, cell seeding density, passage range, staining timing, and imaging parameters. The sensitivity of Cell Painting to these variables means that approximately the same protocol can yield incomparable data.

Analysis pipeline standardization: Agreement on feature extraction (CellProfiler remains the standard), normalization approaches using Cytominer, and output formats enable data sharing. The JUMP-CP profiling recipes we use provide a template, but adoption needs to be broader.

Quality control metrics: The field needs consensus quality control metrics specific to morphological profiling — not just image quality but profile reproducibility, positive control separation, and batch effect magnitude. Poor data should be automatically flagged before contaminating downstream analysis. Single-cell quality control is often overlooked and segmentation errors, debris, and cells touching the edges of images should all be culled from datasets before downstream analysis. The methods for detecting quality issues are maturing rapidly and are converging on standardized approaches.

Validation against clinical outcomes: Ultimately, the value proposition for pharma is predictive accuracy of clinical outcomes. We need prospective studies demonstrating that Cell Painting signatures predict clinical toxicity or efficacy better than alternatives.

Looking ahead, how do you envision the evolution of Cell Painting assays in the next decade, particularly with advancements in automation and data analytics?

The convergence of Cell Painting with AI will define the next era.

Deep learning representation: We are moving away from hand-engineered features that measure predefined shape, intensity, and texture, toward learned representations where neural networks extract optimal features directly from raw pixels, such as DeepProfiler and DinoV3. Early results suggest these models capture subtle biological nuances that traditional feature sets often miss.

Predictive toxicology: Models trained on massive Cell Painting databases, linked to clinical outcome data, will increasingly predict human toxicity before animal or human testing. This aligns with regulatory movement toward new approach methodologies (NAMs) and will transform how safety is assessed early in development. This will improve drug safety overall for clinical trial participants and for patients after drugs get FDA approved. Additionally, if NAMs can reduce failure rates in drug development, they could substantially lower the cost of bringing new drugs to market, helping to make medicines more affordable.

Automated target deconvolution: Matching an unknown compound’s phenotype to a database of known mechanisms will become a standard, powerful approach — using learned embeddings to infer its mechanism of action instead of relying on years of traditional biochemical characterization.

Closed-loop experimentation: I anticipate fully automated systems where AI analyzes screening results and independently designs follow-up experiments — selecting compounds for confirmation, adjusting dose ranges, and scheduling orthogonal assays — with minimal human intervention.

Integration with spatial technologies: Combining Cell Painting with spatial transcriptomics and imaging mass cytometry, technologies my lab is actively developing, will provide unprecedented resolution on how drugs alter tissue architecture and cell-cell interactions in complex in vitro model systems like patient-derived organoids.

Where do you want to see your research go in the next few years?

My goal is to close the gap between throughput and physiological prediction. High-content screening has historically traded relevance for scale where immortalized or cancer cell lines in 2D enable large-scale screens but poorly predict human outcomes.

Over the next few years, my team is working to establish Cell Painting in patient-derived organoid systems as a validated, regulatory-ready platform. For DILI specifically, this means demonstrating that morphological signatures in liver organoids predict clinical hepatotoxicity patterns with accuracy that influences drug development decisions.

The technical components — high-content imaging with our Yokogawa systems, CellPose and CellProfiler segmentation and feature extraction pipelines, and ML classifiers — have matured to the point where we can accurately predict clinical outcomes.

The frontier is validation: showing prospectively that these platforms predict clinical outcomes better than existing in vitro and in vivo approaches. That's the work that will make Cell Painting not just useful for interrogating cellular systems in the laboratory, but actually transformative for producing safer, more effective therapeutics to patients.