Integrating single-cell and bulk transcriptomic perturbation resources reveals complementary therapeutic spaces for drug repurposing

Finding a new cure for a disease is often a slow, expensive, and uncertain journey. Scientists typically spend over a decade and billions of dollars trying to discover a single new medicine from scratch, only to see most candidates fail in late-stage testing. Because of this, many researchers have turned to a different strategy: drug repurposing. This approach involves taking medicines that are already approved and safe for one condition and testing them for entirely different diseases. The logic behind this method relies on a biological principle called transcriptional reversal. Every disease leaves a specific fingerprint on the genes inside our cells, turning some genes on and others off in a pattern that makes us sick. If a drug can trigger the exact opposite pattern—turning the sick genes off and the healthy ones back on—it has the potential to restore the body to a normal state. For years, scientists have used massive databases of gene activity to hunt for these matching drugs, but they have faced a significant hurdle: the data they rely on often comes from a single type of experiment that misses crucial details.

A team of researchers at the University of California, San Francisco, and Stanford University has developed a new way to solve this problem by combining two very different types of genetic data. They created a unified system, which they named CDRPipe, to compare disease fingerprints against drug effects from two distinct sources. The first source is a long-standing collection of experiments using a technology called microarrays, which measure the average gene activity of a whole group of cells mixed together. The second source is a much newer, massive library of single-cell data, where scientists look at the gene activity of individual cells one by one. By merging these two perspectives, the researchers were able to see a much clearer picture of which drugs might work. Their work suggests that relying on just one type of data leaves out a huge portion of potential cures, and that the best way forward is to use both.

The researchers applied their new system to 233 different disease signatures, ranging from common conditions like asthma and diabetes to complex disorders like endometriosis and various cancers. They tested these disease patterns against thousands of drug profiles from both the older microarray database and the newer single-cell database. The results showed that the two databases were not just repeating the same information; they were revealing completely different sets of potential treatments. In fact, the drugs found by the single-cell database overlapped with those found by the older database in only about 3.5% of cases. This means that if a scientist had used only one of these resources, they would have missed nearly all the unique candidates discovered by the other. The single-cell approach proved particularly powerful, recovering known effective treatments for diseases at a much higher rate than the older method. For example, when looking at autoimmune diseases, the single-cell data successfully identified known therapies for nearly 77% of the conditions, whereas the older method found them for only about 21%.

The study also highlighted how the two methods see the world differently. The older database, with its mix of cells, tended to find drugs that target receptors on the surface of cells, which is common for older, well-established medicines like pain relievers and steroids. The newer single-cell database, with its ability to see individual cells, was better at finding drugs that target enzymes inside the cell, such as those used in modern cancer treatments and targeted therapies for inflammation. When the researchers looked at specific diseases, the differences became even more clear. In a case study on endometriosis, a painful condition affecting the uterus, the single-cell approach identified drugs that target the cancer-like growth properties of the disease, such as specific enzyme inhibitors. The older approach, meanwhile, found drugs that manage symptoms, like hormonal treatments and anti-inflammatories. Both sets of findings were valuable, but they pointed to different biological mechanisms. The researchers found that when a drug was predicted to work by both methods, it was a much stronger candidate for success, suggesting that agreement between these two different technologies is a sign of high confidence.

This work does not claim to have solved the problem of finding cures, but it provides a much more reliable map for the journey ahead. The authors emphasize that their findings are based on computational predictions and need to be tested in real biological models and clinical trials. However, the study demonstrates that the field of drug repurposing has been limited by looking through a single lens. By integrating the broad, historical view of older data with the detailed, high-resolution view of modern single-cell technology, scientists can now cast a wider net. This integrated approach allows them to recover known treatments they might have otherwise missed and to identify new, high-confidence candidates that could be tested in patients. The researchers have made their tools and data publicly available, inviting the global scientific community to use this combined perspective to accelerate the discovery of new therapies for diseases that currently have few or no effective treatments.