NEW YORK (GenomeWeb) – Freenome, another new entrant to the circulating tumor DNA analysis market, said last week that it has raised $5.5 million in seed funding to support commercialization of its technology.
Unlike current liquid biopsy tests on the market or in development, Freenome's testing will be based on genome-wide sequencing and a computational strategy the firm calls an "Adaptive Genomics Engine" (AGE).
Gabriel Otte, CEO and cofounder of the company, told GenomeWeb that Freenome's AGE is in essence a computational system that can be trained, and then continually refines its ability to identify patterns in genomic data that differentiate, for example, cancer from not-cancer or one type of cancer from another.
Current commercial liquid biopsy tests, like most tissue-based tests, rely either on the identification of specific known biomarkers, or on the detection of known or novel somatic mutations in a subset of genes most frequently mutated in cancer, or in some cases in the whole exome. But Freenome argues that all of these strategies discard the vast majority of genomic data that might hold the keys to identifying cancer at its earliest inception.
Though he declined to provide details on how the company perfoms it’s genome-wide sequencing and trains the algorithm, Otte said he and his colleagues believe that their unique approach can make possible analysis of genome-wide sequencing data at a practical price point. And though they haven't shared data publicly yet, he said that it looks like it is working.
The strategy is this: perform genome-wide sequencing on samples of healthy individuals and cancer patients and feed the data into the AGE until it learns how to distinguish them blindly. Moving forward, the more samples it sees, the better it becomes at sorting samples into the cancer versus no-cancer buckets.
Once it is powerful enough to be used as a clinical test, Freenome would then be able to sequence circulating cell-free DNA from customers and classify them with high accuracy.
"It's really about things that machines can see and keep track of that humans can't," Otte said. "For example, if you are calling mutations, the way it's done normally is, reads pile up in a certain location. A certain number read 'A' and a certain number read 'G' in the same position. If the reference genome is A and there is a significant number of Gs — past the cutoff point you have defined — you say there is a mutation there."
“But since it's an arbitrary cutoff, if it doesn’t hit the threshold, the entire set of data gets thrown out … even if it's just one read less than what was required.”
“That makes sense … that's what humans can understand,” Otte said. “But machines can keep track of a lot more things, so whether it's one or 1,000 reads that say 'G', it keeps track of that."
If a particular alteration in the data is a sequencing error, the learning engine will eventually filter it out after seeing enough samples because the random nature of the errors eventually marks them, he said.
“If it’s a random error, very few datasets will have it, but if it's biological, it will happen in multiple samples,” Otte explained.
He said the strategy has been successful so far. "After 20 samples that the machine trained on, we had about 85 percent sensitivity and specificity. After about 100, we had close to 98 percent." However, the new company has not yet made public any of the data supporting these numbers.
Once a basic cancer versus no cancer prediction is set, Otte said, new samples can then be used to establish different questions with different buckets, based on the AGE's ability to look simultaneously at oncogenic mutations and at things like nucleosome positioning or other epigenetic markers that distinguish DNA from different sites of origin in the body.
For example, Otte said the company recently reached a milestone in successfully combining prostate and lung cancer into one test, in which the AGE can discern not only the presence of cancer, but also whether it is a lung or a prostate tumor.
Eventually, the company intends to create a pan-cancer test, but it is currently developing tests for one cancer at a time, and then working on integrating them.
Other research groups and companies have also mentioned using epigenetic signatures like nucleosome positioning to track circulating cell-free DNA to a particular tissue of origin.
Also, another young company, Molecular Stethosocope, is developing a test to detect diseases other than cancer by assaying circulating RNA that can be linked to particular sites in the body.
Several other approaches have also been described in the literature, including one using cell type-specific methylation signatures to trace the origin of circulating cell-free DNA, and another that inferred nucleosome patterns from sequenced cell-free DNA that could be linked to a particular tissue of origin.
Freenome appears to be the first, however, to combine early cancer detection with this method of defining tissue specificity.
According to Otte, Freenome’s AGE also has potential for learning from samples of patients who did or did not respond to certain drugs.
“Because we are doing genome-wide sequencing, we can build a new learning engine to look at who responded and who didn't to certain drugs,” he said, something the team is now investigating will collaborators at the University of California, San Diego.
Various other academic collaborators are working with the company to validate novel signatures that the AGE uncovers and incorporates into its predictive strategy. They are supplying blinded samples to confirm the sensitivity of the tests that the company has already developed for prostate, lung, and colorectal cancer.
Otte said that Freenome has already had discussions with the US Food and Drug Administration to make sure it is conducting its development and validations in a way that will support afuture submission of its test for approval as an IVD.
But Freenome may also launch its test as an LDT first to make it clinically available, he said, as it moves down the longer path to an eventual IVD.
The company plans to publish data on the early performance of its AGE-derived cancer detection tests in the near future, he added.