Integrating disease and protein networks expands genetic support for therapeutic targets
The paper introduces BRIDGE, a method that propagates genetic evidence through coupled protein-protein interaction and disease-similarity networks to significantly expand the coverage and clinical success rate of therapeutic target selection beyond what is possible with direct genetic support alone.
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 This is an AI-generated explanation of a preprint that has not been peer-reviewed.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
Finding a new medicine is a gamble of staggering proportions. It takes over a decade and billions of dollars to move a single drug from a laboratory idea to a pharmacy shelf, and most attempts fail. For decades, scientists have looked to human genetics for a way to tilt the odds in their favor. The logic is straightforward: if a specific gene is naturally linked to a disease in people, then a drug that targets that gene's protein might work as a treatment. This approach has become a trusted compass, guiding researchers toward targets that are more likely to succeed in clinical trials. However, this compass has a blind spot. It only points to targets where the genetic link is direct and obvious. In the vast landscape of human biology, many promising targets and diseases lack this direct genetic signature, leaving them off the map entirely.
A new study introduces a method called BRIDGE, designed to expand the reach of this genetic compass without losing its accuracy. The researchers realized that biology does not operate in isolation. Genes interact with one another in complex networks, much like people in a community, and diseases often share underlying biological similarities. By weaving these connections together, the team created a system that can trace genetic evidence from known targets to unknown ones, and from one disease to a related one. They tested this system against a massive database of thousands of drug development attempts. The results show that by connecting the dots between proteins and diseases, scientists can identify many more promising drug targets that were previously invisible, while still maintaining a high likelihood of clinical success. This approach does not replace the need for direct genetic proof, but it offers a way to find valuable leads in the shadows where direct evidence is missing.
The core challenge the researchers addressed is that direct genetic support is rare. In a review of over 13,000 drug development efforts that had reached at least the first phase of human testing, only about 5% of the target-disease pairs had direct genetic evidence linking them. For the vast majority of candidates, the genetic signal was silent. The team asked whether they could borrow strength from the surrounding biological context. They built a digital map that combines two types of networks. The first is a protein interaction map, which shows which proteins physically touch or work together inside a cell. The second is a disease similarity map, which groups illnesses that share biological features or symptoms. These two maps are linked together, creating a vast web where a gene, a protein, and a disease are all connected.
Using this web, the researchers developed an algorithm that acts like a signal booster. It starts with the genes that have strong, direct genetic support for a specific disease. The algorithm then sends a "signal" out through the network. This signal travels along protein connections to neighboring genes and jumps across disease connections to related conditions. If a gene is close to a known successful target, or if a disease is similar to one with a known genetic cause, the algorithm gives that gene-disease pair a higher score. Crucially, the system includes a strict calibration step. It compares the scores of real candidates against scores generated by random shuffling to ensure that the results are not just a product of how well-connected a gene happens to be. This prevents the system from simply favoring popular genes that are connected to everything else.
When the team applied this method to their database of 13,000 drug development pairs, the results were significant. By combining the direct genetic evidence with the new network-derived support, the coverage of supported targets nearly doubled, rising from 5.1% to 9.5%. This means that for nearly 600 additional drug development pairs, the system could now provide a genetic rationale that was previously absent. More importantly, this expansion did not come at the cost of quality. The new candidates identified by the network were just as likely to succeed in clinical trials as those with direct genetic support. In fact, among the pairs that lacked direct genetic evidence but were supported by the network, nearly 25% had successfully reached the market as launched drugs, compared to less than 9% of the unsupported candidates.
The study also demonstrated that this method works even for genes that have no direct genetic history at all. The researchers identified 101 genes that had never been linked to any disease by direct genetic evidence but were still supported by the network because of their connections to other successful targets. These genes were not just theoretical possibilities; many of them were already being pursued as drug targets, and some had already reached the market. The system successfully highlighted these candidates, showing that the network approach could uncover hidden potential in genes that traditional methods would have ignored.
To understand how this expansion happens, the researchers broke down the pathways the signal took. They found two main routes. The first route connects a target to a related disease. For example, if a gene is known to help treat one type of heart disease, the system can suggest it might also work for a similar heart condition, even if there is no direct genetic proof for that second condition yet. The second route connects a target to a neighboring protein. If a drug works on a specific protein, the system can suggest that a protein physically touching it might also be a good target. The study showed that the most effective results came from using both routes together. When the researchers tested the system by removing one of these connections, the number of successful predictions dropped significantly, proving that the combination of disease relationships and protein connections is essential for the method's success.
The researchers also compared their sophisticated network method against simpler rules that scientists might use, such as just looking at immediate neighbors. While simple rules could find some good candidates, the BRIDGE system was better at distinguishing the truly promising ones from the rest. When the researchers ranked the candidates by their network scores, they found more successful drug launches than with the simpler rules. This suggests that the network provides a deeper, more nuanced understanding of the biological relationships that simple proximity cannot capture. The system is able to prioritize candidates that are slightly further away in the network but still biologically relevant, filtering out noise that might mislead a simpler approach.
The study included a forward-looking check to see if the method could predict future successes. The researchers used data from before 2024 to build their model and then looked at what happened to drug development programs between 2024 and 2025. The network-supported candidates continued to show a higher rate of success compared to those without support. Although the number of new launches in this short window was small, the trend held true, suggesting that the method is robust and not just a reflection of past data. The researchers noted that while the sample size for these future predictions was limited, the direction of the results was consistent with their earlier findings.
Two specific examples illustrate how this works in practice. In one case, the system linked a gene involved in a rare metabolic disorder to a broader category of liver diseases. The genetic evidence for the specific disorder pointed to a protein that was physically connected to another protein, which was the target of an existing, successful therapy. The network correctly identified this connection, and the therapy was indeed effective for the broader condition. In another case, the system connected a receptor protein to a viral infection. The receptor was two steps away in the protein network from a gene known to be involved in the disease. Despite this distance, the network signal was strong enough to highlight the receptor as a viable target. Subsequent clinical trials confirmed that drugs targeting this receptor could reduce hospitalizations for the infection, validating the network's prediction.
The researchers acknowledge that this method is not a magic bullet. It relies on the quality of the underlying maps of protein interactions and disease similarities, which are still being refined. There are also limitations in how well the system can handle diseases that are very different from one another or proteins that have few connections. However, the study provides a clear demonstration that integrating different types of biological data can significantly expand the toolkit for drug discovery. It shows that by looking at the broader biological context, scientists can find new paths to effective treatments that were previously hidden.
Ultimately, this work offers a practical way to make the most of human genetic data. Instead of waiting for a direct genetic link to appear for every single disease, researchers can now use the web of biological connections to find leads that are just as promising. This approach does not replace the need for rigorous testing, but it provides a stronger foundation for deciding which drugs to pursue. By expanding the pool of genetically supported targets, the method increases the chances of finding new cures for diseases that currently have few options. The study suggests that the future of drug discovery lies not just in finding the single right gene, but in understanding the complex network of relationships that genes, proteins, and diseases share.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.