In the inaugural research article, we briefly introduced how we 'digitize and decipher’ your samples at Viome using metatranscriptomics and artificial intelligence (AI). Have you ever wondered how a little tube with your sample gets transformed from a biological specimen to your personalized health insights and recommendations on the Viome App and eventually materializes as physical packages of supplements made precisely for you? And exactly how unique are your precision supplements? This article will walk you through the fascinating journey of the metamorphosis of your sample data.
Recall that your DNA code is something like a dictionary of words and phrases for the language of your biology. But the actual story of your life is expressed dynamically in sentences, paragraphs, and chapters by your RNA, by selecting words and phrases that are activated by your environment (including food) and stringing them together to form the story. This dynamic genetic expression of words and phrases is unique to you and holds within the secrets of your illness and your health. It turns out that molecules in your sample contain those expressed words and phrases that make up your unique story. Once we decipher your biological story up to this point, we can create a precise customized nutrition plan for you that can help you write much improved chapters of your life’s story going forward.
Meet ViOS, Viome’s digitizing, deciphering, custom manufacturing, and precision nutrition platform. The ViOS process first turns the molecules in your biological samples (raw materials) into digital information, then analyzes that information using a suite of advanced algorithms to understand your biological pathways, then transforms those personal pathway insights into precise nutritional ingredients made specifically for your consumption. ViOS cajoles your samples to reveal meaningful biological information, sifts through the complex interplay of pathways captured in that information, and creates a set of precision-crafted nutritional ingredients designed to take you to a state of better health.
The diagram below shows the main building blocks of ViOS and an overview of your sample data journey. Let’s walk through this together step by step.
Gathering the elements of your biological narrative
At the beginning of this journey (step 1 in the figure), you, the Viome customer (artfully shown as a stick figure), receive the kit, register it on the App, answer the questionnaires, and send in your sample(s).
Your biological samples (stool, blood, and saliva) are the raw materials for the ViOS process. You will collect your biological samples at home (step 2 in the figure). You can follow the easy-to-follow instructions to prepare and collect your samples based on the kit you purchased. When done, you can send them to the Viome CLIA-certified laboratory.
The answers to your questionnaires that you provide us present a portrait of your experience. In addition to the standard demographic questions, we ask you to tell us about your medical history, symptoms & conditions, oral hygiene, skin health, mental health, sleep habits, digestive health, and your adherence to Viome nutrition recommendations if you are a returning customer. ViOS also has several clinically validated health questionnaires, which are triggered if your answers meet certain criteria for specific health issues. These health questionnaires include clinically validated questionnaires that were designed to assess the symptom severity of issues related to your gut health, sleep quality, physical activity, etc, and physical activity (IPAQ). It may sound like a lot, but you can take your time completing them. And be assured that all information that goes to ViOS is protected.
Together, the samples and the questionnaires capture your phenotype – your observable traits, which outline the elements of your biological narrative. Once we have received your sample(s) and you have completed your questionnaires, you will soon receive your Viome health scores and recommendations. But what exactly does Viome do with your samples and your questionnaire answers? Read on to find out.
Assembling the fragments of your story
Once the Viome Lab receives your biological sample(s), it processes the samples to extract the essence of your biochemistry. The lab first extracts RNA molecules using several innovative lab techniques that were honed over a decade to get meaningful signals from your samples. One of the complexities of this step is that there are a lot of insignificant molecules in the sample that need to be carefully removed, and Viome has developed a suite of techniques for doing so 12. The processed samples are then sequenced using a state-of-the-art sequencer, currently an Illumina Novaseq 6000.
The output of the sequencing step (step 3 in the figure) is a vast amount of sequencing data in fastq format, shown above. Every four lines represent one sequencing read, indicated by the red box. A sample with 10 million reads will generate a file with 40 million lines. A typical sample generates several millions of such sequencing reads – words in your biological story, but all scrambled up. The only way this raw data can be understood is through sophisticated bioinformatics algorithms that can reverse engineer and re-assemble these scrambled words (sequencing reads) to identify the phrases (genes) and books (genomes). The ultimate goal of this step and the following step is to get high-quality data that corresponds to the signals we care about – meaningful information about your biological story.
Here is an analogy to illustrate what the bioinformatics step does. There is a library (=your sample), in which there are a large number of books (=microbes). The library has many copies of some books and fewer copies of others. Each book contains many phrases (=genes). You come into the library and read words from books, which corresponds to those genes being expressed (=your life’s activities). Viome only sees a snapshot of you reading at some time (=sequencing), and we would like to estimate how often each word is being read in the library (=gene expression) or the number of copies of each book (=strain/species abundance), and to understand what the words mean in their context (=gene function). In this way, bioinformatics solves your intricate molecular puzzle and assembles the fragments of your story.
The bioinformatics step ultimately generates (step 4 in the figure) the counts of expressed genes, how much is expressed for each gene, their functions, and how abundant each microbe is in your sample. The pipeline translates the vast amount of sequencing reads to the meaningful molecular features for both the microbiome and the human host. From tens of millions of sequencing reads, the pipelines identify tens of thousands of microbial strains/species, microbial gene functions (in KO), and expressed human genes in samples and report how much of each (abundance) is detected.
Such an accomplishment is made possible with Viome’s ‘genome catalog’, which includes tens of thousands of microbial genomes from NCBI RefSeq release 205 ‘complete genome’ category, many thousands from representative human gut genomes of UHGG 3, ribosomal RNA (rRNA) gene sequences, and the human genome GRCh38 4. These genomes cover archaea, bacteria, fungi, protozoa, phages, viruses, and the human host, with almost 100 million annotated genes.
Unraveling your molecular storylines
The Viome Scoring Engine consumes the genes and genomes identified by the bioinformatics algorithms (step 4 in the figure) to decipher molecular pathway insights. It does this by using a cornucopia of scientific knowledge combined with a number of models of biology developed by training algorithms to detect patterns, also known as machine learning (a subfield of artificial intelligence).
At its root, a health score represents the quantified level of activity of a specific biological pathway. Given Viome’s metatranscriptomics technology, it is possible to quantify the activity of a biological pathway if we understand the components and interactions within that pathway, and we can model them collectively (subject of another blog soon). All the health scores in your Viome App are generated from curated scientific knowledge and machine learning (ML) models within the Viome Scoring Engine. Health score models compute functional scores (between 1 and 100) of various metabolic pathways. These ML models are pre-trained on a Viome Reference Cohort of adult individuals (>=18 yr) with matched stool, blood, and saliva samples. This reference cohort is representative of the Viome adult customer population, with all three sample types and good quality measures in both demographic and molecular data. The health score models compute the scores for you as an individual by comparing your sample profile against the reference cohort.
The biological aging model predicts your biological age, the markers of age represented by patterns in your biological samples, which is distinct from your chronological age 5. For an incoming new sample such as yours, the models predict the respective health scores or biological age by comparing your sample profile against the reference cohort. Please note that at the moment, people younger than 18 do not get a biological age prediction.
An example of the health scores is the ‘butyrate production pathways’ model, which, as suggested by the name, evaluates all microbial activities contributing to the production of butyrate, a short-chain fatty acid known to beneficially affect many wellness areas from gut lining to insulin sensitivity and satiety (feeling full). A score that is not optimal means that your microbial butyrate production could really use a good boost from supplements or foods that either feed or add butyrate-producing microbes into your gut ecosystem.
How does the algorithm decide whether your particular health score is optimal or not optimal? The figure below illustrates our population statistics-based approach. We first evaluate a given score's distribution within the Viome Reference Cohort. The thresholds for not optimal and optimal, represented in red dotted lines, are determined by the median absolute deviation, a measure of dispersion. In this example, the particular pathway being modeled, such as a short-chain fatty acid, is known to be better with more activity. Thus, for example, a score above 69 is considered optimal (~21% of the population), a score below 48 is considered not optimal (~24% of the population). Everything in between is considered average (~55% of the population). We will have a more detailed blog about Viome health scores soon.
The output of the Viome Scoring Engine (step 5 in the figure) is a series of insightful vignettes that make up the essential storyline of your biology, which is available in your Viome App. These insights are also the basis for creating a custom nutritional plan for you.
Crafting your exclusive nutritional tableau
Now that we have recreated your detailed biological story until this point in time, we’re ready to shape the next chapter of your life’s story. But first, we need to understand the many plot twists that got us here – a key one related to your blood sugar response, also known as glycemic response. The other important elements relate to how your biological pathways (scores) from the previous section (step 5 in the figure) relate to the food you consume. All of these analyses are performed by the Viome Recommendation Engine.
Viome’s Glycemic Response Model 6 is another machine-learned model within ViOS, a key part of the Viome Recommendation Engine. This model’s job is to decide whether a particular food in Viome’s food ontology should be ‘minimize’ or ‘enjoy’ for you based on your gut microbiome. The model is pre-trained on about 1,100 study participants whose food nutrient intake, continuous glycemic response, and gut microbiome activity were studied carefully over weeks. It predicts the probability of having a particular food be minimized. Similar to health scores, it compares your sample to the study participants and predicts the probability for a food to be in the ‘minimize’ category. If the probability is above a predefined threshold, the food is ‘enjoy,’ otherwise the food is ‘minimize’.
The Viome Recommendation Engine takes your Viome health scores, glycemic response prediction, and self-reported phenotypes and runs all these inputs through AI-powered algorithms. The AI algorithms are meticulously developed by a team of nutritional and AI experts to incorporate the latest scientific findings, thoroughly validated by a team of clinical experts to ensure that it generates clinically appropriate recommendations. The algorithms output the food and supplement recommendations targeting your Viome predefined conditions. The food recommendation provides what foods are to avoid, to enjoy, to minimize, and superfood. The supplement recommendation provides a set of ingredients with dosages out of hundreds of available ingredients, including precision prebiotics and probiotics, precision supplements like vitamins and herbal extracts, or a combination of both.
You may wonder “how unique are my recommendations”. Let’s take spinach as an example. Even though it is commonly considered a healthy leafy green, if your ‘oxalate production pathways’ score that assesses all microbial activities contributing to removing oxalate is low, spinach would be on your ‘avoid’ list. By the same token, if your ‘oxalate production pathways’ score is high, then spinach would be on the ‘superfood’ list. We will have a designated blog on why some commonly regarded as healthy food may not be beneficial to you (and your gut microbiome).
To get a sense of the uniqueness of every recommendation on a broader population scale, let us compare the similarity of recommendations of two random people. We can do this by counting the fraction of people that had the same food in the same elements in the superfood, avoid food, and supplement ingredients categories. In the figure below, we took 100,000 random pairs of individuals and compared the number of common foods and ingredients in their personalized recommendations. As you can see, the majority of the pairs have 0 supplement ingredients in common, less than 1 avoid food in common, and less than 2 superfoods in common. This data shows that your foods and supplements are custom-crafted for your specific biological needs.
All your health insights and precision recommendations are delivered to the Viome App (step 6 in the figure). You can take time navigating through, as there is a lot of information about your health and what you can do with food or precision supplements. Viome’s Customer Experience team is always there, ready to answer any questions you may have.
Once the precision ingredients and their dosages are identified, the next step (step 7 in the figure) in your sample data’s journey is precision supplements manufacturing. If you have subscribed to one of the Viome supplement subscription plans, the recommended ingredients and dosages are turned into appropriate capsule sachets and powders. The Viome Precision Manufacturing facility is a highly automated manufacturing and packaging facility that creates your precision nutrients into a custom package (step 8 in the figure) to support your unique scores in precise dosages. Each person’s unique formula of food extracts, minerals, vitamins, amino acids, herbs, probiotics and prebiotics is combined into a custom package made to order.
It is very important to note that each ingredient has unique strengths and weaknesses, and these must be matched with your unique biology. For example, your supplement formula may contain curcumin in an appropriate dosage to reduce oxidative stress; however, if your bile acid production is high, curcumin would not be included.* Similarly, garlic extract may be included in your supplements to support a number of beneficial effects; but if your sulfide gas production is already very high, garlic would not be included.* These concepts are discussed in more detail in our publication about data-driven precision nutrition 7 and will be described in upcoming blogs.
You’ve Got Mail! (step 9 in the figure) This is when you embark on your custom-crafted wellness journey. Your precision supplements should become a part of your nutritional lifestyle. As you adopt this lifestyle, you should start seeing improvements in various aspects of your health, and over time you will see progress in your health scores as well. Keeping your health in top shape is not a one-and-done; it is a habit that you need to incorporate into your everyday life. With that habit, you should see improvements in how you feel.
Retesting: discovering your biology’s rhythm
Your custom nutrition should start adjusting and optimizing your biological pathways to be more attuned to your specific needs. In a few months, you will want to retest your molecular markers (scores) to see how you have done. If you have kept up your nutrition regimen, you should see a number of improvements in your scores. When we surveyed 17,500 Viome customers who have re-tested, 70% said they felt better in at least one of the following areas – digestion, sleep, weight, mood, or energy! We have also started doing controlled analyses of individuals with specific health issues 8, and placebo-controlled randomized studies, and we will write more blogs on this topic. And when you re-test, our recommendations and the precision supplement formula are updated based on your retest and the latest science and discoveries. This is ideally a lifelong process, where you continuously track and keep your health pathways in a good zone.
So there you go, the incredible journey of your data: from a sample collected at home to a pool of RNA molecules, to tens of millions of sequencing reads, to tens of thousands of genes and genomes to capture your microbiome and cellular activities, to many health insights using ML/AI, and to the personalized recommendations. Thus a full circle is completed - you help Viome help you to unfold the mysteries of your health and, more importantly, to keep helping you write better and better chapters in the story of your life!
References
Hatch A, Horne J, Toma R, Twibell B, Somerville K, Pelle B, Canfield K, Genkin M, Banavar G, Perlina A, Messier H, Klitgord N, Vuyisich M. (2019). A Robust Metatranscriptomic Technology for Population-Scale Studies of Diet, Gut Microbiome, and Human Health. International Journal of Genomics, Hindawi.com.
Toma R, Cai Y, Ogundijo O, Hu L, Gline S, Demusaj D, Duval N, Torres P, Camacho F, Banavar G, and Vuyisich M. (2023). A clinically validated human saliva metatranscriptomic test for global systems biology studies. BioTechniques.
Almeida, A. et al. (2021). A unified catalog of 204,938 reference genomes from the human gut microbiome. Nat Biotechnol.
Schneider, V. A. et al. (2017). Evaluation of GRCh38 and de novo haploid genome assemblies demonstrates the enduring quality of the reference assembly. Genome Research.
Gopu V, Cai Y, Krishnan S, Rajagopal S, Camacho F, Toma R, Torres P, Vuyisich M, Perlina A, Banavar G, Tily H. (2020). An accurate aging clock developed from the largest dataset of microbial and human gene expression reveals molecular mechanisms of aging. bioRxiv.
Tily H, Patridge E, Cai Y, Gopu V, Gline S, Genkin M, Lindau H, Sjue A, Slavov I, Perlina A, Klitgord N, Messier H, Vuyisich M, Banavar G. Gut Microbiome Activity Contributes to Prediction of Individual Variation in Glycemic Response in Adults. (2021). Diabetes Therapy.
Janelle Connell, Ryan Toma, Cleo Han-Chen Ho, Nan Shen, Pedro Moura, Tiep Le, Eric Patridge, Grant Antoine, Hilary Keiser, Cristina Julian, Uma Naidoo, Damon Tanton, Guruduth Banavar, Momchilo Vuyisich. (2021). [Data-driven nutrition and clinical outcomes for health conditions]. bioRxiv.
Cristina Julian, Nan Shen, Matthew Molusky, Lan Hu, Vishakh Gopu, Anmol Gorakshakar, Eric Patridge, Grant Antoine, Janelle Connell, Hilary Keiser, Uma Naidoo, Momchilo Vuyisich, Guruduth Banavar. (2023). [Effectiveness of precision supplements on mood in US population]. medRxiv.