UNIPRO enables protein inference and panel expansion across single-cell and spatial transcriptomics

Imagine you are trying to understand a bustling city. You have a massive list of every person's thoughts and plans (their "transcriptome"), which tells you who they are supposed to be. But thoughts don't always match actions. Sometimes a person plans to be a chef but ends up working as a mechanic. To really know what's happening in the city, you need to see what tools they are actually holding (their "proteome"). In biology, proteins are those tools—they do the heavy lifting, send the signals, and build the structures of life. While scientists have gotten really good at reading the "thoughts" (RNA) of individual cells, actually measuring the "tools" (proteins) is like trying to interview every single citizen in a massive city: it's expensive, slow, and often you can only ask a few questions at a time. This leaves researchers with a huge map of thoughts but missing the crucial details of what those cells are actually doing.

Now, imagine you have a super-smart detective who can look at a person's detailed diary (their RNA) and accurately guess exactly what tools they are holding, even if you never saw the tools themselves. That's essentially what this paper introduces. The researchers built a new computer program called UNIPRO. Think of it as a master translator that doesn't just read words; it understands the neighborhood. It knows that people living next to each other in a city often have similar jobs or wear similar uniforms. By combining a cell's internal "thoughts" with its physical location and its neighbors, UNIPRO can predict missing protein data with surprising accuracy. This is a big deal because it allows scientists to take existing maps of cell thoughts and turn them into detailed, protein-rich atlases of tissues, revealing hidden patterns in how our bodies are organized and how diseases like cancer build their defenses.

The Detective with a Neighborhood Map

For a long time, scientists have had two main ways to study cells. One way is to look at the RNA, which is like reading a cell's to-do list. The other is to measure proteins, which are the actual workers doing the job. While we have thousands of maps showing the to-do lists of cells, we have very few maps showing the workers, mostly because measuring proteins in tissue is tricky and expensive. Existing computer programs tried to guess the workers based on the to-do lists, but they treated every cell like a stranger standing alone in a field. They didn't realize that in real life, cells are neighbors. A cell in a tumor doesn't act the same way as a cell in healthy tissue, even if their to-do lists look similar, because their surroundings change everything.

The authors of this paper, led by researchers at Harbin Institute of Technology, realized that to guess proteins correctly, you need to understand the neighborhood. They built UNIPRO, a tool that acts like a detective who draws a map of the city. It doesn't just look at one cell; it looks at how that cell connects to its neighbors and how similar its "thoughts" are to the people nearby.

How UNIPRO Works: The Smart Glue

UNIPRO uses a clever trick called a "graph." Imagine a social network where dots are cells and lines connect them. Usually, you connect dots if they are standing next to each other (spatial neighbors) or if they are thinking the same things (transcriptomic neighbors). UNIPRO does both, but it's smart about it.

Sometimes, the physical map is messy. Maybe there's a gap in the tissue, or a spot where the data is missing. If you only looked at who is standing next to whom, you might connect the wrong people. UNIPRO checks how "crowded" a neighborhood is. If a spot is in a dense, busy part of the tissue, it trusts the physical neighbors more. If a spot is in a sparse or messy area, it trusts the "thought" similarity more. It's like a detective who knows that in a crowded market, you can trust who is standing next to you, but in an empty alley, you should trust who you know by their voice.

Once it has this smart map, UNIPRO doesn't just stop there. It keeps refining the connections. As it learns more about the cells, it adjusts the strength of the lines between them, figuring out which neighbors are truly important and which are just coincidentally close. This allows it to create a "protein-aware" representation of the tissue—a mental model that understands both the cell's identity and its environment.

What They Found: Filling in the Blanks

The researchers tested UNIPRO on a bunch of different datasets, including human lymph nodes, tonsils, and breast cancer tissues. Here is what they discovered:

1. It Works Across Different Sections of Tissue
They tried training the program on one slice of a lymph node and asked it to predict proteins on a completely different, unseen slice of the same tissue. UNIPRO was better at this than any other method. It didn't just guess random numbers; it recreated the specific patterns of proteins that define different parts of the tissue, like the outer shell versus the inner core. This suggests that the program learned the "rules" of how proteins are organized, not just memorized the data.

2. It Reveals the Secrets of Cancer
When they applied UNIPRO to breast cancer tissue, the results were striking. Cancer is messy, and it's hard to tell where the tumor ends and the healthy tissue begins just by looking at RNA. But when they added the proteins UNIPRO predicted, the boundaries became crystal clear. The program identified specific zones: areas where immune cells were gathering to fight, areas where blood vessels were growing, and the exact edge where the tumor was invading healthy tissue. It found markers like PDPN and CD86 that highlighted these boundaries, showing a level of detail that RNA alone missed.

3. It Can "Invent" Missing Proteins
One of the most exciting things UNIPRO can do is "panel expansion." Imagine you have a test that measures 20 proteins, but you wish it measured 100. UNIPRO can look at the 20 proteins you have (plus the RNA data) and accurately predict what the other 80 would look like. They tested this by hiding some proteins in their data and seeing if UNIPRO could guess them back. It did a better job than previous tools, preserving the local patterns of the tissue. This means scientists could potentially use cheaper, smaller tests and still get a full picture of the protein landscape.

4. It Works on Single Cells Too
Even though UNIPRO was built to handle spatial data (where location matters), it works just as well on single-cell data where location isn't known. They tested it on blood cells from different people and found it could identify subtle subtypes of immune cells that other methods missed. For example, it could distinguish between different types of T-cells based on their predicted protein levels, helping to sort out confused groups of cells that looked the same on paper.

Why This Matters

The paper suggests that UNIPRO is a powerful new way to get the most out of the data we already have. Instead of needing expensive new experiments to measure every single protein, we can use this tool to infer them from the RNA data we've already collected. This doesn't mean the predictions are perfect replacements for real measurements—the authors are careful to say these are estimates based on patterns—but they are good enough to reveal biological truths that were previously hidden.

By unifying how we look at single cells and spatial tissues, UNIPRO offers a way to build "protein-informed" maps of the human body. Whether it's understanding how a tumor hides from the immune system or how immune cells organize themselves in a lymph node, this tool helps us see the city of our body not just as a list of thoughts, but as a living, working community.