JUMP Cell Painting chemical dataset: Morphological map of 115,796 compounds in human cells
The JUMP-Cell Painting Consortium has released a comprehensive public dataset containing morphological profiles of 115,796 compounds across 1.53 billion single cells, establishing a robust, reproducible reference resource that accelerates drug discovery by bringing image-based profiling to the scale and utility of genomics.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
In the search for new medicines, scientists often face a problem of scale. They must test thousands of chemical compounds to see which ones might change a cell in a helpful way, such as stopping a disease or killing a cancer. For decades, the most common way to do this was to look for a single, specific reaction, like checking if a drug blocks one particular protein. But cells are complex, and a medicine that works often changes many things at once. To understand these changes, researchers developed a method called "Cell Painting." Instead of looking for just one thing, this technique stains a cell with six different colored dyes, each highlighting a specific part of the cell's internal machinery, such as the nucleus, the energy factories, or the skeleton. By taking a photograph of these glowing parts, scientists can see the cell's overall shape and texture. If a chemical changes the cell's health, the photo will look different. This method produces a detailed portrait of the cell's state, capturing thousands of tiny details that a single test might miss.
For a long time, these portraits were scattered across different laboratories, making it hard to compare results or build a large library of knowledge. A new project has now solved this by creating the largest public library of these cellular portraits ever assembled. The JUMP Cell Painting Consortium, a partnership of eighteen organizations including major pharmaceutical companies and research institutes, worked together to test over 115,000 different chemical compounds. They did this not in one lab, but across twelve different laboratories around the world, using various microscopes and equipment. The goal was to see if this massive, messy, real-world data could still tell a clear story about how chemicals affect human cells.
The researchers focused on a specific type of human bone cancer cell, known as U2OS, which is a standard model for these kinds of studies. They treated these cells with 115,796 unique compounds, mostly at a concentration of 10 micromolar, and took high-resolution photographs of the cells after 48 hours. In total, the project generated more than 43 million images, which were then analyzed by computer software to extract a "profile" for each cell. This profile is a list of thousands of numbers describing the cell's appearance. The team combined data from all twelve sites into a single, unified database containing over 1.5 billion individual cell profiles. This massive dataset is now available for any scientist to use, acting as a reference map for how chemicals change cell behavior.
One of the most important questions the team asked was whether data from twelve different labs could be trusted to work together. Since each lab used different microscopes and had different staff, there was a risk that the differences in the equipment would drown out the actual biological signals. The researchers found that despite these technical variations, the biological signals remained strong. About 15 percent of the tested compounds showed a clear, measurable change in the cells' appearance, meaning they were biologically active. While the signal was slightly weaker when comparing results across different labs compared to results within a single lab, the difference was small enough that the data remained useful. This proved that a global network of labs can successfully combine their work to create a single, powerful resource without losing the ability to detect real biological effects.
The power of this library lies in its ability to reveal hidden connections. The researchers discovered that chemicals which share the same biological target or mechanism of action tend to produce very similar cellular portraits. If two different drugs are designed to block the same protein, the cells they treat will look alike under the microscope, even if the drugs look completely different chemically. This "guilt by association" principle allows scientists to take a new, unknown compound and compare its cellular portrait to the millions of others in the database. If the unknown compound looks like a group of known drugs that target a specific protein, it is highly likely that the new compound works in the same way. The study showed that this method is highly accurate; compounds with the same known target were found to be neighbors in the data far more often than would happen by chance.
The team also used the data to explore the relationship between a chemical's structure and its effect on the cell. They found that while chemicals with similar shapes often produce similar cellular changes, this is not always true. Some chemicals that look almost identical structurally can have completely different effects on the cell, a phenomenon known as an "activity cliff." Conversely, some chemicals with very different structures can produce the same cellular outcome. By mapping these relationships, the researchers created a tool that can help drug developers spot these cliffs and avoid chemicals that might cause unwanted side effects. They also found that certain physical properties of a chemical, such as how well it dissolves in fat, make it more likely to cause a visible change in the cell, but these properties do not always predict whether that change will be consistent or specific.
To prove the practical value of this library, the researchers tested it with a real-world challenge. They took a small set of new compounds, known to target a specific receptor involved in metabolism, and treated a different type of bladder cancer cell with them. They then asked the database to find matches for these new compounds without telling the computer what the compounds were. The system successfully identified that these new compounds produced a cellular portrait similar to known drugs that stop cell division. Further experiments confirmed that these new compounds indeed caused the cells to stop dividing and enter a state of aging, just like the known drugs. This demonstrated that a researcher could use this public library to figure out how a new drug works, even if they only have a few samples and are using a different type of cell than the one used to build the library.
The project also highlighted the limits of what this kind of data can do. While the cellular portraits were excellent at predicting how a drug might affect a cell's shape or its immediate targets, they were not good at predicting how a drug would affect a whole human body, such as causing liver damage or heart problems. This is because the test cells are grown in a dish and lack the complex systems of a living organism, like the liver's ability to process drugs. The researchers noted that to predict these broader effects, future studies would need to use different types of cells that better mimic human organs.
Ultimately, this work establishes a new standard for how biological data is shared and used. By bringing together the efforts of twelve laboratories and over 115,000 compounds, the consortium has created a public resource that is as rich and detailed as the genomic databases used to map human DNA. It allows scientists to ask questions about chemical behavior that were previously impossible to answer, turning the visual complexity of a cell into a navigable map for drug discovery. The data is freely available, inviting the global scientific community to explore it, find new connections, and accelerate the development of medicines for diseases that currently have no cure.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.