The Cancer Gene Census (CGC) is an expert-curated catalogue of every gene with strong evidence of driving human cancer, maintained as part of the larger COSMIC (Catalogue of Somatic Mutations in Cancer) database. It serves as a shared reference standard across basic research, clinical diagnostics, and drug development, giving scientists and clinicians a common list of genes whose mutations are known to push cells toward malignancy. As of its major 2018 expansion, the CGC described 719 cancer-driving genes in detail, with a growing second tier of 145 additional candidates supported by newer but less complete evidence. The resource matters because cancer genomics generates enormous amounts of data, and without a rigorously vetted list of which genes actually matter, distinguishing a true cancer driver from the thousands of harmless bystander mutations in any tumor would be far more difficult.
What the Census Actually Contains
The CGC is not simply a list of gene names. For each of its entries, expert curators compile functional and mechanistic descriptions of how mutations in that gene contribute to cancer. This includes which of the recognized hallmarks of cancer the gene affects, whether the mutations cause a gain of function (turning a gene into an accelerator) or a loss of function (disabling a brake), and what impact those changes have on the resulting protein. The information comes from published peer-reviewed literature, and every entry must meet a high evidentiary bar before inclusion.
The census uses a two-tier system. Tier 1 genes have extensive, well-documented evidence from multiple independent studies confirming their role in cancer. Tier 2 genes come from more recent research and show promising but less detailed indications of involvement. This tiered approach lets the census grow with the science without diluting the confidence level of its core entries. Tier 2 acts as a kind of holding area where newer candidate genes can accumulate evidence over time before potentially graduating to the first tier.
Why a Curated List Matters More Than You Might Think
A typical tumor carries thousands of mutations. Most are “passengers,” random changes that happened during the rapid cell division but do not contribute to the cancer itself. Only a handful are “drivers,” the mutations that actually gave the tumor its growth advantage. Telling the two apart is one of the central challenges in cancer genomics, and the CGC provides the anchor for that effort. When a sequencing lab finds a mutation in a patient’s tumor, one of the first things they check is whether that gene appears in the CGC. If it does, the mutation is far more likely to be clinically meaningful.
This matters at every stage. In basic research, the CGC helps scientists design experiments around genes with established cancer roles rather than chasing false leads. In pharmaceutical development, it helps companies identify the most promising drug targets. And in clinical oncology, it informs which mutations get flagged in a patient’s genomic report, potentially guiding treatment decisions about targeted therapies or clinical trial eligibility.
How the Census Fits Among Other Cancer Knowledge Bases
The CGC is not the only database cataloguing cancer-relevant genetic information. A comparative analysis of major precision oncology knowledge bases, including the CGC, OncoKB, CIViC, and several others, found that while these resources largely overlap in the genes they cover, each one also contains a notable amount of unique information. Some databases focus more on gene-drug associations, while others, like the CGC, emphasize broader biological characterization of how genes contribute to cancer.
What sets the CGC apart is its scope and its focus on curation depth. Rather than simply listing which mutations are “actionable” for a specific drug, the CGC aims to describe the full landscape of genetic dysfunction across all human cancers. This makes it especially useful as a reference standard. When researchers develop new computational tools to predict cancer drivers, for example, they routinely benchmark their results against the CGC to see how many known drivers their method successfully identifies. A pan-cancer analysis of 33 cancer types using 26 different computational tools found 299 driver genes, of which 59 had not previously been reported by earlier studies or in the CGC, illustrating both the census’s breadth and the fact that new drivers continue to be discovered beyond its borders.
Inherited Versus Acquired Mutations
Most people hear “cancer gene” and think of inherited risk, like the BRCA1 mutations associated with breast and ovarian cancer. But the majority of genes in the CGC are catalogued primarily for their role in somatic mutations, the ones that arise during a person’s lifetime rather than being inherited from a parent. The distinction matters because these two categories of mutations often work through different mechanisms and have different implications for patients and their families.
That said, the boundary between inherited and acquired cancer genes is blurrier than it once seemed. Research has shown that some genes previously known only for their somatic role in cancer also harbor rare inherited variants that predispose people to the disease. A study examining this overlap found that tumor suppressor genes, which normally act as brakes on cell growth, showed a significant enrichment of rare damaging inherited variants in patients whose tumors had also lost the remaining working copy of that gene. This pattern is consistent with the classic “two-hit” model of cancer, where one defective copy is inherited and the second is knocked out during life. Genes that act as accelerators (oncogenes) did not show this same pattern, which makes biological sense since oncogenes typically need to be switched on rather than lost.
The practical takeaway is that the CGC’s catalogue of somatic cancer genes is also a valuable starting point for discovering new inherited cancer predisposition genes. Researchers can use it to systematically search for inherited variants that might increase cancer risk, potentially expanding the list of genes that genetic counselors screen for in high-risk families.
Most Cancer Genes Are Tissue-Specific
One of the most persistent misconceptions about cancer genetics is that a “cancer gene” drives cancer everywhere in the body. In reality, the evidence points in the opposite direction. A large-scale analysis of 571 cancer genes across 265 cancer types found that tissue specificity is the norm, not the exception. Of all the genes examined, only 5 showed broad enrichment across more than half of the cancer types studied. The vast majority, including deeply studied genes like BRCA1 and KRAS, were significantly enriched in only a smaller subset of cancer types.
This has real consequences for how you should think about cancer risk and treatment. KRAS mutations, for instance, are extremely common in pancreatic cancer and certain lung cancers but play a much smaller role in many other tumor types. Calling KRAS a “cancer gene” without specifying the tissue context is a bit like saying aspirin is “a drug” without mentioning what it treats. The CGC captures some of this nuance by documenting which cancer types each gene is associated with, but the tissue specificity finding underscores that even the best catalogue requires context to be used properly.
For clinicians, this means that finding a mutation in a CGC-listed gene does not automatically make it a driver in every tumor. The gene’s known tissue associations have to match the patient’s cancer type before the mutation should be treated as clinically significant. For researchers, it means that pan-cancer analyses, while powerful, can miss drivers that operate in only one or two tissue types if the statistical methods are not sensitive enough to detect tissue-specific signals.
The Challenge of Non-Coding Mutations
The CGC has historically focused on mutations in the protein-coding regions of genes, which make up less than 2% of the human genome. The other 98%, the non-coding regions, contain regulatory elements that control when and how much of each protein gets made. Mutations in these regulatory regions can also drive cancer, but finding them is far harder. The non-coding space is more than 50 times larger than the coding regions, meaning mutations are abundant, yet reliable identification of which non-coding mutations are actually under selection pressure remains a significant challenge.
Emerging research suggests that non-coding driver mutations are relatively infrequent compared to coding drivers, though this could partly reflect the difficulty of detecting them rather than their true rarity. Statistical methods for finding non-coding drivers have to account for the fact that mutation rates vary dramatically across the genome depending on factors like DNA repair activity and chromatin structure. Specialized tools have been developed to handle this complexity, incorporating sample-specific mutational signatures, long-range variation in mutation rates, and position-specific measures of functional impact.
One creative approach uses evolutionary conservation across species. The logic is straightforward: if a stretch of non-coding DNA has been preserved across mammals for millions of years, it is probably doing something important, and mutations there are more likely to be functional. A study of medulloblastoma, a childhood brain cancer, combined whole-genome sequencing data with evolutionary constraint information inferred from 240 mammal species to identify genes enriched in non-coding mutations at conserved positions. This kind of analysis can nominate new candidate driver genes that would be invisible to methods focused only on coding regions.
For the CGC, non-coding drivers represent a frontier. As methods improve and more whole-genome (rather than just protein-coding) sequencing data becomes available, the census will likely need to expand its framework to accommodate regulatory mutations that do not fit neatly into the traditional gene-centric model.
Childhood Cancers and the Limits of Adult-Focused Models
The CGC was built largely on data from adult cancers, and childhood cancers turn out to be genomically quite different. Pediatric tumors generally carry far fewer mutations than adult tumors and are driven by variants that disrupt transcriptional activity, chromatin state, and non-coding regulatory regions rather than the accumulated point mutations typical of adult cancers. This means that a catalogue optimized for the mutational patterns of adult cancers may not fully capture the drivers of childhood malignancies.
Efforts to address this gap are underway. A pediatric-specific gene panel, SJPedPanel, was designed to cover the mutations most relevant to childhood cancers. When tested against a cohort of pediatric patients with known disease-causing variants, the panel covered about 86% of those variants, including roughly 82% of the gene rearrangements responsible for fusion proteins, a type of mutation especially common in childhood leukemias and sarcomas. In a separate validation, the panel detected about 91% of known disease-causing variants at standard sequencing depths, and roughly 95% of variants when the mutant DNA made up at least half a percent of the sample.
These numbers are encouraging, but they also highlight the ongoing need to refine cancer gene catalogues for specific patient populations. A resource built primarily from adult solid tumors will inevitably underrepresent the fusion genes, epigenetic regulators, and developmental pathway disruptors that are central to pediatric oncology.
Combining Rare Variants with Common Risk
The CGC focuses on high-impact mutations in specific genes, but cancer risk is not always driven by a single dramatic mutation. For common cancers like breast and prostate cancer, risk also depends on the cumulative effect of many common genetic variants, each contributing a tiny amount of additional risk. Researchers are now exploring what happens when you combine both layers of information.
A study examining breast and prostate cancers found that people who carried both rare high-impact gene variants and a high polygenic risk score (a measure of accumulated common-variant risk) faced dramatically elevated odds. For breast cancer, the combination produced odds ratios as high as roughly 15-fold compared to people with intermediate common-variant risk and no rare variants. For prostate cancer, the combined effect reached as high as about 22-fold. The rare variants alone increased risk, and the common-variant background alone increased risk, but the two together were more than additive.
This has implications for how the CGC’s gene list gets used in clinical risk assessment. Knowing that a patient carries a rare variant in a CGC-listed gene is important, but understanding their broader genetic background can sharpen the risk estimate considerably. A person with a pathogenic BRCA1 variant and a high polygenic risk score faces a materially different lifetime risk than someone with the same BRCA1 variant but a low polygenic score. Gene panels built around CGC entries are starting to be used alongside polygenic risk scores to produce more layered, individualized risk profiles.
The Evolutionary History of Cancer Driver Genes
Cancer driver genes did not evolve to cause cancer. They are normal genes with essential roles in cell growth, DNA repair, immune signaling, and development that happen to become dangerous when mutated. Researchers have started examining the evolutionary history of these genes to understand why they are vulnerable to cancer-driving mutations in the first place.
A study evaluating the evolution of 568 cancer driver genes across 66 cancer types looked at selection pressures over two timescales: the long arc of primate evolution spanning millions of years and more recent selection in modern human populations over the last hundred thousand years or so. The findings add nuance to how we think about these genes. Some cancer drivers show signs of strong evolutionary conservation, meaning that natural selection has been aggressively preserving their sequences because any change tends to be harmful. Others show evidence of adaptive evolution, suggesting that certain changes in these genes were actually beneficial at some point in human history, even though those same regions can now be exploited by cancer.
Separately, research examining the conservation of specific cancer-driving mutations across vertebrates has found that the top-ranked driver mutations tend to sit in the most highly conserved regions of their respective proteins. These mutations are almost never seen in other species, which makes sense: if a mutation appears frequently in tumors but is essentially absent across the animal kingdom, that is strong evidence that natural selection has been eliminating it for hundreds of millions of years because of how damaging it is. The fact that it keeps appearing in human cancers reflects the mutation’s powerful growth-promoting effect in the short term, precisely the kind of change that evolution would ordinarily purge.
This evolutionary perspective has practical value beyond academic interest. Researchers can use conservation scores to help prioritize which newly discovered mutations are most likely to be true drivers. If a mutation falls in a region that has barely changed across vertebrate evolution, it is a stronger candidate for functional importance than one in a region that varies freely across species. Some newer driver-detection methods explicitly incorporate evolutionary constraint as one of their scoring criteria, and the CGC’s curators can use this type of evidence when evaluating candidate genes for inclusion.
How the Census Keeps Growing
The CGC is not a static document. As sequencing costs drop and large-scale cancer genomics projects generate data from increasingly diverse populations and tumor types, new candidate driver genes continue to emerge. The pan-cancer analysis that identified 299 drivers using 26 computational tools found 59 genes not previously catalogued in the CGC or reported by other major studies, indicating that the current census, while comprehensive, is not complete.
Several forces are pushing the boundaries of what the census will need to cover in the coming years. Whole-genome sequencing, as opposed to the exome-only sequencing that dominated earlier cancer studies, is generating data on non-coding mutations that may require new categories of entries. Pediatric cancers, with their distinct mutational profiles and emphasis on structural variants and epigenetic disruption, demand dedicated attention. And the growing integration of germline and somatic data means that the neat boundary between “inherited cancer predisposition gene” and “somatic cancer driver” will continue to blur, potentially requiring the CGC to annotate genes for both roles more systematically.
The tissue specificity findings also suggest that future expansions of the census should go beyond simply adding new genes. For existing entries, richer annotation of which tissue contexts each gene operates in would make the resource more useful for both researchers running computational analyses and clinicians interpreting patient genomic reports. A CGC entry for KRAS that specifies its strongest associations with pancreatic, colorectal, and lung adenocarcinoma is more clinically actionable than one that simply flags KRAS as an oncogene.