This article has been updated from a previous version to clarify comments made about the original applications for the technology licensed from LANL.
NEW YORK (GenomeWeb) – A little over two years into its existence, direct-to-consumer microbiome company Viome has released a paper that describes aspects of the sequencing protocol and analysis pipeline it uses to provide direct-to-consumer microbiome testing to guide nutritional choices.
In January, the company published a pre-print paper that describes Viomega, the technology on which the company is building its business. The paper includes details on Viome's protocols for collecting and preserving samples; extracting and sequencing RNA; its tools for taxonomic classification of microbes including a database of over 110,000 microbial genomes; and tools for analyzing microbial gene expression and quantifying enzymatic action.
However, researchers who reviewed the paper noted that while the results showcase the reproducibility of Viome's test, it does not offer enough information to fully evaluate the company's claims about its products.
Viome analyzes stool samples to determine the microbial composition of the human gut and claims to use machine learning to make personalized nutritional recommendations that balance the gut microbiome, including what customers could do to minimize the production of harmful metabolites and maximize the production of beneficial ones. In the paper, the company reports the results of analyzing data from 10,000 stool samples collected using its Gut Intelligence kit. It uses its internal informatics pipeline to identify several thousand microbial strains and species in the samples and quantify their biochemical functions. The paper also includes the results of smaller studies on short- and long-term stability of stool samples and intra-sample variability.
Viome opened its doors in 2016. According to Guruduth Banavar, the company’s chief technology officer, Founder and CEO Naveen Jain saw an opportunity to take advantage of technology being developed at Los Alamos National Laboratory to aid in understanding environmental biome diversity and potential pathogens. "There's all of this investigation and discoveries that have been made as to the behavior of microbes," Banavar said in an interview. "[And] this technology was proposed by one of the Los Alamos scientists as a possible technology that could be used for something in the world of commercial healthcare."
The company's interest in microbial technologies coincided with increasing research and funding for studies focused on exploring the microbiome's composition and associations with various conditions. Banavar was working for IBM at the time but became interested in applying his expertise in artificial intelligence to perceived gaps in healthcare. "I join[ed] this effort to create a product that could give people visibility into what is going on in the microbiome, … what can be done to improve the composition of the microbiome through normal lifestyle changes, and what impact that will have on chronic diseases," he said.
After licensing the technology from LANL, the company has put together a product for analyzing human gut microbiomes, which it now is commercializing. It has also raised $20.5 million in funding in two rounds from investors such as Khosla Ventures and Physician Partners.
The first component of the product is a home delivery kit mailed directly to customers to collect and submit stool samples for analysis. The package includes a simple collection tube that contains a chemical denaturant and RNA stabilizing solution that preserves the RNA molecules in the sample for a few weeks. Customers ship their sample back in a prepaid envelope to Viome's lab in Los Alamos. Once Viome receives the kit, the company processes and extracts RNA from the sample and sequences it. It analyzes the resulting reads using a custom bioinformatics pipeline that identifies taxa and gene signatures in the data. Full details of these protocols are provided in the pre-print.
"We look at which of these taxa and genes are active and, based on activity levels, we can determine which pathways are enabled and active," Banavar explained. And, based on the activity of those pathways, "we determine what is likely going on with your microbiome." For example, "we can say whether you have specific viruses and whether you have good metabolism." The company also claims that it can combine information gleaned from the microbiome with data about compounds in food to come up with recommendations about which foods will most likely be beneficial to customers.
Some of this information is collected through Viome's app which the company uses to deliver the results of its testing. In the results section, the company provides aggregate scores and graphs that offer a picture of the customer's internal biology, similar to the sorts of test results that individuals receive when physicians run routine clinical testing. The app also provides recommendations for dietary adjustments, lifestyle changes, and food supplements to improve gut flora based on test results and information from existing resources.
The app includes a questionnaire that customers fill out detailing information such as their dietary habits, how much they sleep, and how often they drink alcohol. The questionnaire also collects data on recurring symptoms like insomnia and migraines as well as information on current medications for diabetes, high cholesterol, and so on. "There is a huge amount of benefit that you can get just by looking at the foods and how you modulate the foods and supplements that you eat in your daily life," Banavar said.
Two different teams review the information that goes into Viome's app. The first is a team of translational scientists with expertise in microbiology and pathway interpretation who carefully review and validate the output of Viome's pipelines. Many of their recommendations have been implemented in the company's analysis software. "We use a lot of their domain knowledge in the algorithms that we have implemented [in the pipeline] that runs automatically," Banavar said. Furthermore, the company has implemented a validation procedure that ensures that its pipeline is using the most up-to-date information to generate the scores and graphs that it reports in the results.
A second team comprised of clinicians and nutritional experts validate the food recommendations that the company provides to customers. Much like the microbiologists who provide domain expertise for the company's analysis algorithms, "the clinicians give us the domain expertise [that we] incorporate into the recommendation engine," Banavar said.
Viome has tested the efficacy of its recommendations on validation cohorts, but its clinicians also double check the output of the engine to ensure that it is delivering the right recommendations for the right reasons. Also, where the company's AI infrastructure is unable to interpret customers' data, the system refers these cases to Viome's teams of translational scientists and clinicians for manual review. The results of the human analyses is then fed back in to the company's software so that it learns what to do when it faces ambiguous cases.
Customers receive the results of their Viome analysis within two to three weeks although it can take up to four weeks because it takes time to get enough samples back in the mail for the sequencing step. "If we were to scale it up to where there are, say, 1,000 samples that arrive every day in our lab, we will probably reduce our turnaround time dramatically because we don't have to wait to fill up our entire assay," Banavar said. "We can probably turn it around in less than two weeks."
Viome opted for a direct-to-consumer business model, Banavar said, so that it could assess the real-world impact of the recommendations that it makes. Since the product launched officially in April last year, the company has mailed out tens of thousands of kits. The current price point for Viome's test, according to its website, is $299 down from $399 last year.
So far much of the evidence for the efficacy of Viome's product seems at least in part anecdotal. According to Banavar, the company has gotten feedback from customers who report improvements in various health conditions ranging from obesity to irritable bowel syndrome that they attribute to changes made based on Viome's recommendations.
Last month, the company entered into an agreement with Campbell Soup Company to acquire Habit, a personalized nutrition company. According to Viome, the acquisition will expand the company's ability to provide personalized nutrition plan recommendations and engagement tools for customers. The company is also enrolling people in internal studies including one focused on studying the blood transcriptome in older patients, and another that is focused on personalizing diets to minimize individual blood sugar responses. Past studies have looked at functional changes in the gut microbiome and biomarkers of chronic inflammation.
Weighing the evidence
There is a lot of skepticism surrounding DTC microbiome tests and DTC genetic testing in general, and some of Viome's critics have been very vocal about their distrust of the company, arguing that it has not provided peer-reviewed scientific evidence to back claims that it can make personalized nutrition recommendations to customers that maximize the production of beneficial metabolites and minimize harmful ones.
"Until they actually publish some data and present some results this is no different from Theranos," Jonathan Eisen, a professor at the University of California, Davis and director of the UC Davis Microbiome Special Research Program, said in an interview. Eisen previously served on the scientific advisory board for Viome competitor uBiome and currently advises a company that does animal microbiome testing. He did not review the Viome paper for this article.
"Microbiome diagnostics is interesting and there's lots of people doing it" but usually "you present data before telling people you can do a diagnostic service," he said. In terms of the way it makes personalized nutrition recommendations, Viome is "not presenting any evidence that anything they do has a scientific justification behind it … there is something really off."
Researchers who reviewed the Viome paper noted that although the evidence there seems to support claims regarding the reproducibility of the company's test, it is lacking on the same critical details that have been sticking points for the larger microbiome community — questions about exactly how the company makes microbiome-based food recommendations including details about algorithms used.
A researcher from a prominent academic institution conducting a large microbiome metatranscriptomics study who did not have permission from his organization to speak on the record and requested anonymity argued that the paper does not provide any new insights.
"What they [Viome] clearly demonstrated is that they are able to do metatranscriptomics," the researcher said. "They can isolate RNA and perform reproducible metatranscriptomics and they have a software pipeline that allows them to use metatranscriptomics to profile microbial communities to the strain level." However, "almost any sequencing center could do metatranscriptomics" and "there are also publicly available versions of software that do the same thing so that's not really surprising that this company could do these things," the researcher noted.
Furthermore, the company has claimed to provide "personalized dietary advice based on metatranscriptomics but did not show any of the derivations for those personalized methods," the researcher added. "They show extraordinarily generic metrics — for example the top 100 [Kyoto Encyclopedia of Genes and Genomes] pathways — [but] no information on how they are using this to derive personalized dietary recommendations."
The researcher also noted that the company's claim that its metatranscriptomics process can identify all regular microorganisms in the gut is at least somewhat misleading because the company's approach, at least as described in the paper, does not suggest that samples are enriched for virus-like particles.
Elisabeth Bik, a microbiologist, science consultant, and founder of the Microbiome Digest blog, noted that the study shows that the Viome's method detects a range of microorganisms with relatively equal efficiency and precision. Bik previously worked for uBiome but has since left the company. "It is important to first point out that this paper is not a peer-reviewed paper," she said in an email. "It is a first step, however, to share the work that Viome did to build a metatranscriptomics platform and show some of their first results."
She pointed to experiments reported in the paper where the company tested stool from a few participants, varying the sampling sites, time from sample collection to testing, and the length of time between sample collection. It presents the results of its testing in terms of variation in microbial composition and gene function composition. "In all three experiments, samples from the same stool specimen or individual were very similar to each other, showing that the Viome test results are reproducible," she said.
However, "it was interesting to see that in all cases the microbial composition was better able to tell individuals apart than the gene expression," she noted. "That is a bit ironic, because the preprint states at the beginning … that functional gene analysis, not microbial composition, is needed for personalized health insights." However the paper "appears to show that each person has their own, personal gut microbiome, while the functional capacities appear to be pretty similar between individuals."
Bik also noted that although the paper claims that metatranscriptomics is superior to other microbial analysis techniques, the results presented in the paper do not support this point. "There was no comparison of 16S sequencing versus metatranscriptomics, and the amount of viruses, archaea, and eukaryotes in the sample set was not very high" which suggests "that 16S sequencing did not miss as much diversity as the authors claim in the introduction," she said. For example, "figure 4 [in the paper] is a colorful representation of experiments but is lacking percentages. There is also no word on how the negative controls did perform over their tests of 10,000 samples."
Bik also pointed out that the company offers little information on it analysis of 10,000 samples though these are mentioned in the title. "It was disappointing to see so few results on the data. This dataset, which is one of the largest of its kind, sounded very promising, and I had hoped to see many more figures on these samples," she said. "The authors presented a table with the prevalence of microbial taxa and functional genes in this sample set, but did not do any additional analysis on relative abundance or correlation of microbial taxa or genes with lifestyle or diet. I am not sure why this dataset is part of the paper at all."
Finally, Bik noted that the paper does not answer questions about how "small person-to-person variations in microbial gene expression will lead to the 'goal to develop personalized nutrition algorithms' as stated in the abstract." Furthermore, she added, the paper does not offer a lot of technical detail about the company's protocols for sample extraction or sequencing library preparation, which would be required for the paper to pass peer review.