More Than Half of DNMT3A-Mutant Patients in Pooled Public Cancer Genomic Data Carry No Tumor-Derived Mutation: Clonal Hematopoiesis Introduced by Patient-Keyed Aggregation
This study reveals that a standard patient-keyed aggregation of public cancer genomic data erroneously classifies clonal hematopoiesis mutations from blood normals as tumor-derived mutations in over half of DNMT3A-mutant patients, creating significant pan-cancer distortions that can be corrected by filtering out blood-derived calls prior to matrix construction.
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 This is an AI-generated explanation of a preprint that has not been peer-reviewed.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
Cancer researchers have long relied on a powerful method: gathering genetic data from thousands of patients across many different hospitals and studies into one massive, shared library. By pooling this information, scientists can spot patterns that are too rare to see in a single hospital, such as which gene changes drive a specific type of cancer or how those changes affect a patient's survival. To make this work, computers link records together using a simple rule: if two samples belong to the same person, they are treated as one entry. This approach has become the standard for understanding cancer on a global scale. However, this method assumes that every piece of genetic data in the library comes from the tumor itself. In reality, patients often have other genetic changes in their blood that are unrelated to their cancer, known as clonal hematopoiesis. These are natural changes that occur in blood cells as people age, and while they are harmless in the blood, they can look exactly like cancer mutations if the computer cannot tell the difference between a blood sample and a tumor sample.
A team of researchers at Stanford University discovered that this confusion has quietly distorted a large portion of the world's public cancer data. They found that when scientists merge different studies into a single database, the computer sometimes mistakes a blood sample for a tumor sample. This happens because three specific studies from Memorial Sloan Kettering Cancer Center deposited blood samples from patients whose tumors were already recorded in other studies. Since the blood and tumor samples shared the same patient ID, the computer merged them together. The result was that harmless blood mutations were incorrectly labeled as cancer mutations in the final database. The researchers traced over one million genetic calls across 483 studies and found that while the error affected only a small fraction of the total data, it was not spread evenly. Instead, it concentrated heavily on specific genes that are known to change in aging blood cells.
The most affected gene was DNMT3A. In the uncorrected database, more than half of the patients flagged as having a mutation in this gene actually had no mutation in their tumor at all; the only evidence came from their blood. For other genes like TET2 and CHEK2, the situation was similar, with roughly one-fifth of the flagged patients having no tumor-derived mutation. In contrast, well-known cancer drivers like TP53 and KRAS were barely touched by this error, which is why many researchers had not noticed the problem until now. The researchers confirmed the nature of the mistake by looking at the details of the genetic changes. The mistaken blood mutations appeared at very low levels, often less than one percent of the genetic material, whereas true tumor mutations usually appear at much higher levels. Furthermore, the patients with these mistaken blood-only mutations were significantly older, matching the pattern of aging blood rather than the typical age of cancer diagnosis.
This error did not stay isolated to the blood studies; it spread into 65 other research groups, almost all of which were studying solid tumors like lung, liver, or stomach cancer. In some of these groups, more than one in ten patients carried a false mutation flag. The distortion was so strong that it made it look like certain cancers had much higher rates of specific mutations than they actually did. For example, in one group of lung cancer patients, the rate of DNMT3A mutations appeared to be nearly five times higher than in another group of lung cancer patients, simply because the first group happened to include patients whose blood samples had been mixed in. Once the researchers removed the blood-derived data, the rates between the two groups became nearly identical, proving that the difference was an artifact of the data collection, not a real biological difference.
The impact of this mistake went beyond just counting mutations; it also changed the conclusions about how these genes affect patient survival. In some cases, the error made a gene look like it was helping patients live longer, when in reality it had no effect. In other cases, it hid a real protective effect. The researchers showed that simply removing the blood data corrected these survival estimates, moving them in ways that random chance could not explain. They provided a simple set of rules for other scientists to follow: before combining studies, check if a genetic call comes from a blood sample or a tumor sample, and if it comes from a blood study, exclude it from the tumor analysis. This fix is straightforward and can be applied to existing databases without needing to re-sequence any patients.
The researchers were careful to note that their correction does not solve every problem. It removes the errors caused by mixing blood and tumor data in the database, but it cannot remove blood cells that have physically invaded a tumor sample and are present inside the tumor tissue itself. That is a separate biological phenomenon that requires different tools to detect. However, the error they identified was purely a result of how the data was put together. By fixing this specific mistake, the scientific community can ensure that the massive libraries of cancer data they rely on reflect the true biology of tumors, rather than a confusing mix of blood and cancer. The work serves as a reminder that even when data is collected correctly, the way it is combined can create new errors, and that keeping track of where every piece of information came from is essential for finding the truth.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.