Dog DNA tests have gotten remarkably good at what they were built to do. Pet parents send a cheek swab, they wait for a few weeks, and then they get a report listing any inherited conditions their dog may carry a genetic risk for.
Often, these risks include bleeding disorders, drug sensitivities, eye diseases, conditions caused by specific mutations that researchers have identified and catalogued over decades of study. For many dogs, that information is genuinely useful.
But these tests come with a limitation. Most dog parents do not know about it.
The diseases that often cause death to dogs, including cancers and cardiac conditions that cut short otherwise healthy lives, are not caused by the kinds of mutations those tests are looking for.
Which means a clean result on a conventional genetic health screen can sit in a dog’s file for years while the actual risk goes undetected. To help pet parents get a better idea, we have to explain what health risk markers show, and what is their limitation.
What Is a Health Risk Marker?
A health risk marker is a specific DNA variant, a mutation, that researchers have identified as being linked to an inherited disease or a higher probability of developing one.
The best way to describe it is a known trouble spot. Scientists and researchers have spent years, studying thousands of dogs. They have identified certain genetic mutations, and traced those mutations to specific conditions.
Those connections now form a catalog. When you test your dog’s DNA is tested, the lab compares the genetic profile of your dog against that catalog. If they find a mutation from the list, they flag it. If they do not find any mutation from the catalogue, you get a clean result.
The test returns one of three results: clear, carrier, or at risk. That last category means the dog carries a mutation researchers have connected to a specific condition, not that the disease is guaranteed to develop.
For a real range of conditions, this approach is genuinely useful. A dog flagged for MDR1 sensitivity changes which medications a vet will prescribe. A bleeding disorder marker changes how a surgeon plans a procedure. When pet parents find a single known mutation that can cause a specific condition, they can change something concrete in their dogs’ life.
Where the Approach Runs Into a Wall
The cancers that kill dogs most often, including lymphoma, hemangiosarcoma, osteosarcoma, mast cell tumors, and glioma, do not trace back to a single mutation. Neither do most cardiac conditions that shorten lives. Each develops through the accumulation of many small genetic contributions, none of which is decisive on its own. Risk builds the way pressure builds, slowly, from many directions at once.
That is what polygenic means in practice. Not one answer, one location, or a single mutation. Instead, hundreds of small signals distributed across the genome, most of them individually unremarkable.
Researchers have spent years trying to find the single genetic source behind these cancers. What they keep finding instead is that there is no single source. A study examining bone cancer across 492 dogs from three different breeds identified 33 separate locations in the genome that each contribute to disease risk. Even with all 33 mapped out, scientists still could not account for the full picture of which dogs got sick.
A separate study tracking blood cancers in golden retrievers found two key risk locations in the genome, but together those two locations explained only about 20 percent of total cancer risk. The other 80 percent remains scattered across the rest of the genome, in patterns too spread out for conventional testing to read.
A third study, published in PLOS Genetics, confirmed the same reality for another cancer type: the predisposition is not one mutation doing the damage, but many smaller risk factors adding up across multiple chromosomes.
Here is the practical implication for pet parents. When a cancer risk is spread across dozens of genetic signals, how can you believe a test that searches only for known, pre-listed mutations? This test will not flag many risks. And that is not because the test is failing, but because this kind of risk was never in its catalog. Simply put, the test was not designed to assess the risk outside of a catalogue of known mutations.
A clean result on a conventional health DNA test does not rule out risk for these diseases. It simply says that none of the mutations on that predefined list were found. That is a different statement from “this dog is not at risk.”
How SmartDX Approaches the Problem
Most DNA tests work like a keyword search. They look for specific words, a known mutations, that researchers have already identified as warning signs. SmartDX works differently. Instead of scanning for keywords, it reads the whole document and asks whether the overall pattern looks like the documents that came from dogs that got sick.
SmartDX analyzes more than 120,000 genetic markers distributed across the entire canine genome at once. Rather than checking a list of pre-selected variants, it applies machine learning to evaluate complex patterns of genetic variation across the full genome. SmartDX does not ask the question “does this dog carry a specific mutation?” Instead, it asks the question “does this dog’s genomic profile resemble the profiles of dogs that developed this disease?”
And pet parents get a different answer.
This is the kind of problem machine learning was built for. Not because it is smarter than traditional analysis, but because it can hold more in view at once. A researcher scanning for specific mutations looks at individual words. A machine learning model trained on thousands of genetic profiles reads the whole sentence and asks whether it sounds like a dog that got sick.
Research published in Communications Medicine in 2023 found that training deep neural networks to recognize distributed risk patterns, rather than individual markers, substantially outperformed standard statistical methods for predicting polygenic diseases in humans. The same principle applies for canines, where the most dangerous conditions are exactly the ones that leave no single footprint.
The Technology Behind It
SmartDX’s process runs through two proprietary systems.
Morse handles the preprocessing step. It takes raw, unformatted data from DNA sequencing, and converts it into a clean, standardized format that the machine learning model can work with. In plain English, Morse handles the cleanup step between a messy raw recording and the final audio file.
Morse functions as the bridge between raw sequencing output and the modeling stage. It collects samples are collected using DNA Genotek Performagene kits and sequences them on the Thermo Fisher Axiom Canine HD array.
Mew is the machine learning engine that does the actual prediction. It has been trained on genomic data from thousands of dogs with confirmed health histories, dogs that developed specific diseases, and dogs verified to have reached old age without developing them.
From that training data, Mew has learned to recognize genomic patterns that appear in dogs that eventually got sick, and patterns that appear in dogs that stayed healthy. When it evaluates a new dog’s genome, it compares that dog’s full genomic profile against those learned patterns and returns a lifetime risk prediction.
Training and Validation
The training data spans more than 300 canine breeds, which reduces the risk of the model performing well for some populations and poorly for others. To address a common constraint in veterinary genomics, that fully labeled datasets are difficult to build at scale, SmartDX applied semi-supervised learning techniques to more than 14,000 additional unlabeled canine genomes.
Model Performance
Cancer Panel (test set, n = 510)
| Disease | Accuracy | Sensitivity | Specificity |
| Lymphoma | 0.91 | 0.86 | 0.94 |
| Hemangiosarcoma | 0.87 | 0.93 | 0.87 |
| Canine Mammary Tumor | 0.95 | 0.97 | 0.95 |
| Mast Cell Tumor | 0.92 | 0.85 | 0.93 |
| Glioma | 0.99 | 0.76 | 0.99 |
| Osteosarcoma | 0.91 | 1.00 | 0.90 |
Accuracy reflects the model’s overall correct rate. Sensitivity is how reliably it catches dogs that do carry elevated risk. Specificity is how reliably it clears dogs that do not.
Cardiac Panel (test set, n = 30 per disease)
| Disease | Accuracy | Sensitivity | Specificity |
| Arrhythmogenic Ventricular Cardiomyopathy | 1.00 | 1.00 | 1.00 |
| Tricuspid Valve Disease | 1.00 | 1.00 | 1.00 |
| Dilated Cardiomyopathy | 1.00 | 1.00 | 1.00 |
The cardiac panel numbers come from a smaller test group of 30 dogs per condition. The smaller validation sets tend to produce cleaner-looking numbers. The cardiac results are promising, but less extensively validated than the cancer panel’s 510-dog test set.
What the Results Mean
SmartDX returns results in three categories: At Risk, Not At Risk, and Uncertain.
At Risk means Mew’s model predicts that the dog’s genomic profile is consistent with dogs that developed the disease during their natural lifetime. This is a prediction(LINK TO ARTICLE 4 Predictive vs Diagnostic TESTS HERE), not a current diagnosis. The dog may or may not be showing any symptoms.
Not At Risk means the model predicts the dog’s profile is not associated with lifetime development of that specific condition, within the diseases the panel covers.
Uncertain means the prediction falls at the boundary between risk categories. The genomic signal is close enough that assigning a definitive result in either direction would not be accurate. Rather than force a result, the test returns an uncertain classification.
The Difference That Can Prolong Your Dog’s Life
Conventional health marker tests and SmartDX both use DNA. They are not answering the same question.
Conventional tests identify whether a dog carries specific mutations that researchers have already linked to known conditions. They are reliable for inherited diseases with identifiable causal variants, and for those diseases, they provide clinically meaningful information.
SmartDX estimates the likelihood that a dog will develop specific cancers or cardiac conditions over their natural lifetime, by evaluating how more than 120,000 genomic markers combine into a pattern, not by checking a list of known mutations. The diseases it covers are the ones responsible for the highest rates of serious illness and shortened lifespans in adult dogs, and they are precisely the ones that conventional mutation-based screening was not designed to reach.
What a SmartDX result gives a pet parent is an earlier starting point. A dog flagged at risk for hemangiosarcoma is a dog whose vet now has a reason to run bloodwork more frequently, to look harder at subtle changes in energy or behavior, to have a conversation that does not usually happen until something is already wrong. That is not a guarantee. But it is a different conversation, at a different point in the timeline.
The two tests are not competing with each other. A conventional marker panel and a SmartDX genomic assessment cover different ground. For many dogs, running both produces a more complete picture than either one alone can provide.