Viome Uses RNA Sequencing, AI to Predict and Target Chronic Disease
Ai4
Analyzing dynamic RNA expression across human and microbial pathways enables personalized nutritional interventions that resolve chronic symptoms far more effectively than static DNA testing.
## Background
Published on YouTube by Ai4 on August 28, 2025, this talk features Guru Banavar, Founding CTO of Viome Life Sciences. The description outlines how Viome combines peta-scale microbial and human gene expression data with AI/ML to detect molecular pathway disruptions early and deliver personalized interventions against chronic disease.
## TL;DR
Guru Banavar presents Viome's multi-omics health platform, which analyzes dynamic host and microbiome RNA from stool, blood, and saliva to track systemic health rather than relying on static DNA. Processing peta-scale biological datasets through AI models, the platform forecasts glycemic responses, calculates biological age, and stratifies risk across conditions like cardiovascular disease and IBS. Viome uses these pathway insights to generate individualized dietary recommendations and custom formulations, backed by randomized controlled trials and employer pilot outcomes. Banavar concludes by detailing how these solutions scale across direct-to-consumer kits, enterprise wellness programs, and clinical-grade provider tools.
## Atmosphere & style
The presentation carries a polished, authoritative, and data-driven keynote tone characteristic of high-level biotechnology conferences. Banavar speaks with calm, executive confidence, methodically translating complex concepts in transcriptomics, bioinformatics, and systems biology into accessible operational frameworks. The mood is measured yet distinctly visionary, creating a feeling of scientific rigor applied directly to practical preventive medicine.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=4s) Viome focuses on proactive chronic disease prevention and lifespan extension by shifting from traditional drug development to dynamic RNA sequencing of stool, blood, and saliva.
- [00:04](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=4s) Unlike static DNA, dynamic RNA captures real-time gene expression (where only ~5% of genes are active at any time) as influenced by diet, sleep, stress, and lifestyle factors.
- [04:31](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=271s) The human microbiome functions as a biochemical sieve with approximately 100 microbial genes for every human gene, directly modulating host immunity and metabolic regulation.
- [07:13](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=433s) Viome has trained over 30 production disease models on peta-scale biological datasets exceeding 10^15 sequenced RNA nucleotides and 10^16 longitudinal phenotype data points.
- [07:13](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=433s) The platform's AI/ML suite accurately predicts individual postprandial glycemic responses, calculates biological versus chronological age, and identifies oral and throat cancer biomarkers.
- [12:48](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=768s) Disease risk stratification identifies precise underlying metabolic pathways rather than treating broad diagnostic labels, such as isolating methane gas or TMA synthesis in heart disease and IBS.
- [15:50](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=950s) Algorithmic precision nutrition dynamically adjusts dietary advice based on pathway function, such as demoting nutrient-dense foods like spinach to 'Avoid' if oxalate elimination is impaired.
- [15:50](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=950s) Placebo-controlled clinical trials showed that personalized nutrition and supplement interventions resolved IBS constipation symptoms in 64% of active participants versus 10% in the placebo group.
- [18:09](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1089s) A six-month enterprise pilot program yielded weight reduction in 58% of participants and improved energy levels in 100% of participants, driving measurable workplace productivity gains.
- [18:44](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1124s) The molecular health ecosystem operates across three primary deployment channels: direct-to-consumer health kits, employer wellness programs, and the clinical-grade Precision Health Pro test for medical practitioners.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=4s) Introduction to Viome and the Multi-Omics Platform
Guru Banavar introduces Viome's distinct approach within the healthcare landscape, shifting the focus from traditional drug development to a consumer-accessible and clinically actionable platform designed to prevent chronic disease and improve lifespan. Since entering the market in 2017, Viome has scaled globally across more than 100 countries, collecting approximately one million non-invasive samples—specifically stool, blood, and saliva—from half a million individuals.
```mermaid
flowchart LR
accTitle: Viome Translational Flywheel
accDescription: Flywheel showing how consumer data feeds AI models and translates back into clinical insights and personalized interventions.
subgraph DataCollection["Data Ingestion"]
A["At-Home Samples<br/>(Stool, Blood, Saliva)"] --> B["Global Cohort<br/>(500k+ Users, 100+ Countries)"]
end
subgraph AIPlatform["AI & Science Core"]
B --> C["RNA Sequencing &<br/>Phenotype Profiling"]
C --> D["Continuous AI/ML<br/>Model Updates"]
end
subgraph Translation["Interventions & Insights"]
D --> E["Personalized Nutrition &<br/>Clinical Grade Tests"]
E -.->|"Flywheel Feedback"| A
end
```
Rather than focusing on DNA, proteins, or standalone metabolites, Viome strategically chose RNA as its foundational molecule of interest. While DNA is static and remains largely unchanged from birth until long after death, RNA is dynamic—only about 5% of genes are actively expressed at any given point in time. Although RNA is notoriously unstable, Viome developed proprietary stabilization techniques enabling high-throughput metatranscriptomic and human host transcriptome sequencing from at-home samples. This data feeds into a continuous learning AI/ML architecture that constantly refines its models as new phenotypes and longitudinal samples arrive, creating a translational R&D flywheel that spans direct-to-consumer and professional clinical-grade diagnostic tests.
- Founded in market in 2017; user base spans over 100 countries.
- Ingested ~1,000,000 stool, blood, and saliva samples from 500,000+ individuals.
- Anchored on three foundational technologies: RNA technologies, microbiome sequencing, and AI/ML modeling.
---
[04:31](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=271s) RNA and the Human-Microbiome Molecular Dance
RNA acts as the dynamic biological bridge translating genetic predisposition into real-time molecular execution, influenced constantly by diet, sleep patterns, physical activity, and stress. Viome emphasizes the critical need to intercept this transcribed messenger RNA to uncover how environmental and lifestyle inputs perturb metabolic pathways and trigger disease states.
A central driver of these pathways is the human microbiome. While there is roughly one microbial cell for every human cell in the body (concentrated heavily in the gut, oral cavity, and skin), the microbial genetic potential dwarfs the human genome: for every single human gene, there are approximately 100 microbial genes. When food passes through the gut, the microbiome functions as a biochemical sieve, fermenting and transforming dietary inputs into active gut metabolites that modulate systemic immunity, metabolic regulation, and organ function. Viome's scientific core models this continuous "dance" and bidirectional communication between human host gene expression and microbial gene expression.
- Ratio of microbial to human genes in the body: ~100 to 1.
- Cellular composition: at least 1 microbial cell for every 1 human cell.
- Microbiome serves as an active metabolic filter that directly dictates immune response and systemic chronic disease risks.
---
[07:13](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=433s) Peta-Scale Biological Data and Predictive AI Models
Operating at the intersection of high-throughput sequencing and artificial intelligence, Viome has processed real-world data scaling to over $10^{15}$ (peta-scale) sequenced RNA nucleotides. When cross-referenced with dense longitudinal phenotype metadata—including medical history, diagnoses, clinical symptoms, medications, dietary habits, and demographic profiles—the total biological dataset exceeds $10^{16}$ data points.
```mermaid
flowchart TD
accTitle: Viome AI Prediction and Recommendation Architecture
accDescription: Flowchart illustrating data processing from raw multi-omics features through AI models to clinical scores and nutritional recommendations.
subgraph InputFeatures["Multi-Omic Inputs & Metadata"]
F1["Salivary & Gut Metatranscriptome"]
F2["Whole Blood Host Transcriptome"]
F3["Phenotype & Medication Metadata"]
end
subgraph ModelTypes["AI/ML Predictive Models"]
M1["Glycemic Response Predictor"]
M2["Biological Age Engine"]
M3["Pathway Scores<br/>(Inflammation, TMA, Oxalate)"]
M4["Cancer Diagnostic Classifiers<br/>(Oral & Throat Cancer)"]
end
subgraph OutputEngine["Clinical & Consumer Translation"]
R1["Food Recommendation Engine<br/>(Enjoy / Minimize / Avoid)"]
R2["Disease Risk Stratification<br/>(20+ Chronic Conditions)"]
R3["Precision Supplement Formulations"]
end
InputFeatures --> ModelTypes
ModelTypes --> OutputEngine
```
Viome has deployed over 30 production disease models alongside hundreds of exploratory research models powering its clinical pipeline and mobile consumer applications. Key predictive capabilities include:
- **Glycemic Response:** Machine learning algorithms mapped across a comprehensive food ontology predict an individual's specific postprandial glucose spike (flat vs. extreme spikes) for hundreds of ingredients.
- **Biological Age vs. Chronological Age:** Stool, blood, and saliva transcriptomes are projected against population-wide age-stratified molecular cohorts. For instance, a 32-year-old whose molecular patterns match a 38-year-old cohort receives an actionable warning signal.
- **Pathway and Phenotypic Predictions:** Models accurately predict sleep quality, PPI and antibiotic history, C-section birth history, cognitive function pathways, oral dental pathways, and gut inflammatory activity.
- **Diagnostic Oncology Classifiers:** Salivary metatranscriptome diagnostic classifiers distinguish patients with oral and throat cancers from healthy cohorts.
- **Advanced Machine Learning Toolset:** Employs bioinformatics pipelines, molecular pathway models, molecular foundation models, and generative LLMs for translating complex molecular scores into intuitive patient-facing insights.
---
[12:48](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=768s) Cellular Pathways, Health Zones, and Disease Risk Stratification
The platform aggregates molecular and microbial pathway outputs into standardized "health zones"—including Gut, Oral, Cardiovascular, Kidney/Heart, and Mental Health—measuring total systemic inflammation, biological age, and pathway-level dysregulation. Viome provides risk profiles for roughly 20 distinct chronic conditions, such as Sjögren's syndrome, ADHD, periodontal disease, and cardiovascular disease.
```mermaid
flowchart LR
accTitle: Suboptimal Pathway Convergence in Heart Disease Risk
accDescription: Example of three dysregulated metabolic pathways contributing to elevated cardiovascular risk.
subgraph DysregulatedPathways["Suboptimal Pathway Status"]
P1["Elevated LDL Cholesterol Pathway"]
P2["High Methane Gas Production"]
P3["Elevated TMA Production Pathway"]
end
DysregulatedPathways ==> DR["Flagged Risk:<br/>Cardiovascular / Heart Disease"]
```
Rather than treating diagnostic labels or symptoms uniformly, Viome identifies the exact underlying pathway mechanism:
- **Cardiovascular Disease Example:** An elevated risk score may be driven by a specific combination of attention-zone pathways: elevated LDL cholesterol pathways, microbial methane gas production, and trimethylamine (TMA) synthesis pathways.
- **IBS and Constipation Deconstruction:** Clinical constipation is not treated as a single monolithic disorder; the system differentiates between sulfide gas accumulation, methane gas production, or secondary bile acid imbalances to prescribe targeted corrective action.
---
[15:50](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=950s) Algorithmic Nutrition and Clinical Trial Outcomes
Dietary advice is strictly individualized based on molecular and metabolic pathway interactions. Foods traditionally categorized as universally beneficial are re-evaluated based on specific pathway activity. For example, while spinach is nutrient-dense and has a low glycemic impact (placing it in a baseline "Enjoy" category), if a patient exhibits an impaired oxalate elimination pathway, spinach is automatically demoted to an "Avoid" recommendation to prevent hyperoxaluria and kidney stone formation.
```mermaid
flowchart TD
accTitle: Algorithmic Food Categorization Logic
accDescription: Logic showing how an impaired oxalate elimination pathway demotes a food from Enjoy to Avoid.
A["Food: Spinach"] --> B{"Predict Glycemic<br/>Response"}
B -->|"Low Spike"| C["Initial Tier: Enjoy"]
C --> D{"Evaluate Active Pathways<br/>(e.g., Oxalate Elimination)"}
D -->|"Suboptimal / Impaired"| E["Final Tier: AVOID<br/>(Risk of Oxalate Overload & Kidney Stones)"]
D -->|"Healthy / Optimal"| F["Final Tier: ENJOY"]
```
Viome validates these precision nutrition algorithms and targeted custom supplement formulations via rigorous randomized, double-blind, placebo-controlled clinical trials:
- **Irritable Bowel Syndrome (IBS):** 64% of participants in the active personalized intervention arm achieved complete resolution of constipation symptoms, compared to approximately 10% in the placebo control arm.
- **Anxiety and Diabetes:** Interventions produced statistically and clinically significant improvements across metabolic control markers and anxiety score endpoints over placebo arms.
- **Enterprise Deployments:** Validation is continuously extended into large-scale corporate wellness programs and clinical cohorts to track longitudinal prevention and symptom reversal.
[18:09](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1089s) Real-World Outcomes: Employer Pilot Results and Well-Being Impacts
Guru Banavar highlights the tangible, real-world outcomes observed when Viome's full diagnostic analysis and personalized intervention protocols are deployed in a structured setting. Reviewing the results of a six-month employer pilot program, he shares that 58% of participating employees achieved measurable weight reduction. Even more strikingly, 100% of participants self-reported improved energy levels across the board following the programmatic implementation.
Beyond primary metrics like weight and gut health, the benefits extended deeply into the daily lives of participants. Banavar emphasizes that the improvements were not confined to personal quality of life; employees also saw notable gains in workplace focus and productivity. This dual impact creates a clear win-win scenario, delivering meaningful health improvements for employees while driving better engagement and organizational performance for employers.
- **Weight Loss:** 58% of pilot participants experienced weight reduction over the 6-month program.
- **Energy Levels:** 100% of participants reported noticeable improvements in energy.
- **Holistic Impact:** Enhanced gut health translated into higher day-to-day focus, boosted workplace productivity, and an overall elevated quality of life.
---
[18:44](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1124s) Viome's 8-Year Evolution, Clinical Offerings, and Organizational Implementation
Reflecting on Viome’s broader trajectory, Banavar notes that the company has spent eight years building and refining its science-backed product ecosystem, continuously iterating on its predictive models and molecular insights. Today, these solutions are engineered to serve multiple audiences across distinct access channels:
```mermaid
flowchart TD
accTitle: Viome Platform Deployment Channels
accDescription: Breakdown of Viome's product offerings across direct consumers, corporate employers, and healthcare providers.
subgraph Platform["Viome Molecular Intelligence Platform"]
V["8+ Years of AI & RNA Diagnostics"]
end
subgraph Channels["Deployment Channels"]
DTC["Direct-to-Consumer<br/>(Individual Health Kits)"]
EMP["Enterprise Employers<br/>(Organization-Wide Wellness)"]
PRO["Healthcare Providers<br/>(Precision Health Pro)"]
end
V --> DTC
V --> EMP
V --> PRO
PRO --> Feat["Clinical-Grade Stratification<br/>& Scientific Evidence"]
```
For individual users, Viome’s tests and customized supplements are directly accessible for at-home wellness management. For companies seeking to support their teams, the platform scales into an organization-wide wellness benefit. Finally, for clinicians and medical practitioners, Viome offers a dedicated clinical-grade test known as the **Precision Health Pro** test. This specialized tier equips providers with deep molecular stratification, advanced clinical guidance, and comprehensive scientific backing underlying every insight.
Concluding his presentation, Banavar invites attendees—whether individual practitioners or enterprise leaders—to trial the test and scan the on-screen QR code to explore deploying Viome’s molecular health platform within their own organizations.
## Notable quotes
> Chronic disease is a major epidemic in the 21st century, with more than 50% of the world population suffering some form of chronic disease leading to lowered quality of life and shorter lives. — Guru Banavar
> The vast majority of chronic diseases are due to environmental and lifestyle factors having a major impact on our molecular pathways. — Guru Banavar
> At Viome, our mission is to detect disruptions in those pathways as early as possible, and provide health insights and interventions that are precisely tailored to individuals to maintain a healthy state. — Guru Banavar
> To this end, we have developed a platform that has extracted peta-scale gene expression data from the microbial and human host cells from more than a million non-invasive samples collected at home from individuals in over 100 countries. — Guru Banavar
> We have analyzed this high-dimensional and longitudinal data using AI and ML techniques, discovered a range of host-microbiome biomarkers associated with a range of chronic diseases, derived health insights from these biomarkers, and delivered personalized interventions through a suite of consumer and professional products. — Guru Banavar
> In this talk, I will present how this approach can lead to a new era of molecular health for the global population. — Guru Banavar
> [00:04](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=4s) While DNA is static and remains largely unchanged from birth until long after death, RNA is dynamic—only about 5% of genes are actively expressed at any given point in time. — Guru Banavar
> [04:31](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=271s) While there is roughly one microbial cell for every human cell in the body... for every single human gene, there are approximately 100 microbial genes. — Guru Banavar
## Mentioned
Viome Life Sciences, Guru Banavar, Ai4, Precision Health Pro, RNA, Microbiome, Sjögren's syndrome, ADHD, Irritable Bowel Syndrome (IBS), Trimethylamine (TMA)
## Video description
Guru Banavar, Founding CTO, Viome Life Sciences
Abstracts
Chronic disease is a major epidemic in the 21st century, with more than 50% of the world population suffering some form of chronic disease leading to lowered quality of life and shorter lives. The vast majority of chronic diseases are due to environmental and lifestyle factors having a major impact on our molecular pathways. At Viome, our mission is to detect disruptions in those pathways as early as possible, and provide health insights and interventions that are precisely tailored to individuals to maintain a healthy state. To this end, we have developed a platform that has extracted peta-scale gene expression data from the microbial and human host cells from more than a million non-invasive samples collected at home from individuals in over 100 countries. We have analyzed this high-dimensional and longitudinal data using AI and ML techniques, discovered a range of host-microbiome biomarkers associated with a range of chronic diseases, derived health insights from these biomarkers, and delivered personalized interventions through a suite of consumer and professional products. In this talk, I will present how this approach can lead to a new era of molecular health for the global population.
Viome employs dynamic RNA metatranscriptomics and artificial intelligence to model host-microbiome interactions, aiming to stratify chronic disease risk and provide individualized nutritional interventions.
Dynamic RNA Sequencing and Microbiome Interactions (0:04)
Guru Banavar explains that Viome selected dynamic RNA over static DNA because only roughly 5% of genes are actively expressed at any given time, capturing real-time metabolic shifts driven by diet, stress, and lifestyle. To overcome RNA's natural instability, Viome developed proprietary stabilization methods to conduct high-throughput metatranscriptomic and human host transcriptome sequencing from at-home stool, blood, and saliva samples. Since 2017, the company reports collecting roughly one million samples from over 500,000 individuals across more than 100 countries.
The sequencing captures the interplay between the human host and the microbiome. Although human and microbial cells exist in an approximate 1:1 ratio, microbial genes outnumber human genes roughly 100 to 1. The microbiome acts as a biochemical filter, converting dietary inputs into active metabolites that influence systemic immunity and metabolic regulation. Viome's core technology models this bidirectional communication between host and microbial gene expression to identify early disease perturbations.
Peta-Scale Data and Machine Learning Architecture (7:13)
Viome's predictive pipeline operates on more than 10^15 sequenced RNA nucleotides, expanding to over 10^16 data points when combined with longitudinal phenotypic metadata such as medical history, symptoms, medications, and diet. The company has deployed over 30 production disease models alongside hundreds of exploratory research models, utilizing bioinformatics pipelines, molecular foundation models, and generative large language models to interpret multi-omic data.
The models perform varied predictive tasks, including projecting tissue transcriptomes against population cohorts to calculate biological age versus chronological age, and predicting personalized glycemic responses to specific food ingredients. Banavar also notes the deployment of diagnostic oncology classifiers, which analyze salivary metatranscriptomes to distinguish oral and throat cancer patients from healthy controls.
Pathway-Level Risk Stratification and Precision Nutrition (12:48)
Rather than evaluating clinical symptoms in isolation, Viome organizes molecular data into health zones—such as gut, oral, cardiovascular, and mental health—to generate risk profiles for roughly 20 chronic conditions. For instance, cardiovascular risk is evaluated through convergent pathway signals like elevated LDL cholesterol, trimethylamine (TMA) production, and microbial methane production. Similarly, the system deconstructs symptoms like constipation into specific drivers, differentiating between sulfide gas accumulation, methane production, or secondary bile acid imbalances.
Dietary recommendations are adjusted based on these active metabolic pathways. While foods such as spinach are typically considered healthy and have low glycemic impact, Viome's recommendation engine demotes spinach to an 'Avoid' status if a user exhibits an impaired oxalate elimination pathway, aiming to mitigate the risk of hyperoxaluria and kidney stone formation.
Clinical Trial Outcomes and Commercial Deployment (15:50)
Viome evaluates its precision interventions through randomized, double-blind, placebo-controlled trials. In a trial evaluating irritable bowel syndrome, Banavar reports that 64% of participants receiving active personalized interventions experienced complete resolution of constipation symptoms, compared to approximately 10% in the placebo arm. Statistically significant improvements were also reported for anxiety and glycemic endpoints over placebo controls. In an employer-based six-month pilot, 58% of participants achieved weight reduction, and 100% self-reported improvements in energy levels.
Banavar outlines three primary channels for Viome's offerings: direct-to-consumer health testing kits and customized supplements, enterprise wellness packages for employers, and 'Precision Health Pro,' a clinical-grade diagnostic test designed to provide healthcare practitioners with molecular stratification and scientific evidence.
Viome positions its RNA metatranscriptomic platform as a scalable, data-driven framework for chronic disease prevention and personalized lifestyle intervention. However, broader clinical integration and provider adoption will depend on sustained validation of its proprietary AI models and clinical trial endpoints across independent populations.
## Background
本视频由 Ai4 于 2025 年 8 月 28 日发布在 YouTube 平台。视频记录了 Viome Life Sciences 联合创始人兼首席技术官 Guru Banavar 的主题演讲，页面简介指出，慢性病已成为波及全球半数以上人口的严重流行病，Viome 致力于通过采集全球超百万份非侵入式居家生物样本提取 PB 级微生物与人体 RNA 表达数据，并借助 AI/ML 挖掘生物标志物以提供精准个性化干预，开启全球分子健康新时代。
## TL;DR
Guru Banavar 详述了 Viome 如何依托“动态 RNA、微生物组与人工智能”三大支柱构建从疾病萌芽期进行主动拦截的分子健康平台。平台突破性地押注极易降解但能实时反映生理状态的 RNA，结合宿主与共生微生物的转录组数据，在 $10^{16}$ 维度的海量生物学数据空间中训练出餐后血糖反应、生物学年龄及 30 余种慢性病预测模型。通过分子通路模拟与食物本体库，系统能够纠正大众饮食认知误区并输出个性化膳食与定制补充剂方案，其干预有效性已在 IBS 双盲随机对照临床试验及企业员工健康管理项目中获得显著验证。
## Atmosphere & style
整场演讲兼具深厚的学术严谨性与清晰的产业落地导向，展现出沉稳笃定、逻辑致密的科技前沿峰会氛围。Guru Banavar 采用平实从容的语调娓娓道来，从宏观的被动医疗范式反思，无缝层层拆解至微观分子通路（如草酸盐与三甲胺代谢）和底层 AI 架构。硬核数据指标、清晰的技术飞轮图示与扎实的临床循证结果贯穿始终，让观众能够直观感受到多组学数据与 AI 深度融合在重塑精准预防医学中的巨大变革力量。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=4s) Viome 打造了直接落地的慢性病预防平台，通过全球 100 多个国家的 50 多万名用户收集近 100 万份粪便、血液和唾液样本，构建了极具多样性的大规模人群分子队列。
- [01:37](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=97s) 相较于静态的 DNA，Viome 押注能够动态反映基因开启状态与环境影响的 RNA，并攻克了室温下稳定并保存 RNA 样本的专利技术难题。
- [01:37](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=97s) 平台构建了人源宿主 RNA 与微生物 RNA 动态互作的转化医学飞轮，用户样本持续驱动 AI 算法自迭代，并反哺 DTC 产品与临床诊断工具。
- [04:30](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=270s) 人体共生微生物基因量约为人类自身基因的 100 倍，食物经肠道菌群代谢产生的代谢物深刻影响着全身免疫与慢性疾病的发展进程。
- [07:12](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=432s) Viome 实验室测序的 RNA 序列达到 PB 级（超 10^15），结合详尽临床与生活方式表型数据后，形成了维度高达 10^16 的海量生物学数据集。
- [07:12](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=432s) 基于海量数据，Viome 训练并验证了 30 余种疾病预测模型，涵盖个体化餐后血糖反应预测、基于多组学的生物学年龄评估，以及口腔咽喉癌筛查模型。
- [11:15](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=675s) 先进 AI 架构整合了分子通路模型、个性化推荐引擎、用于生成通俗报告的 LLM 交互层，以及作为通用诊断分类器的分子基础模型。
- [12:48](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=768s) 移动端 App 呈现六大生理健康分区与 20 余种慢病风险，能够将心脏病等风险精准溯源至三甲胺（TMA）生成、甲烷代谢与 LDL 胆固醇等具体分子通路异常。
- [15:19](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=919s) 食物推荐引擎采用双重过滤机制：先依据宏量营养素预测血糖反应初筛，再根据微量营养素对代谢通路（如草酸清除）的影响进行动态降级或个性化补充剂定制。
- [17:06](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1026s) 在针对 IBS 便秘患者的随机双盲安慰剂对照临床试验（RCT）中，Viome 精准干预组的症状完全缓解率达到 64%，远高于安慰剂对照组的 10%。
- [17:06](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1026s) 为期半年的企业健康试点证实，个性化干预使 58% 参与者体重下降，100% 受试者精力提升，并有效增强了员工职场专注度与生产力。
- [18:45](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1125s) Viome 已构建起覆盖 C 端消费者（DTC）、企业员工福利健康计划，以及面向临床医生深度循证赋能的 Precision Health Pro 专业诊断三大商业生态矩阵。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=4s) 慢性病预防新视角：从药物研发转向分子健康平台
在以慢性病预防和改善健康寿命为核心的分享中，Viome 联合创始人兼首席技术官 Guru Banavar 首先将观众的视野引向一个与传统医疗截然不同的维度。不同于大会前序嘉宾 Eric 所聚焦的重资产、长周期的药物研发赛道，Banavar 呈现的是一个已经商业落地并在市场上实际运转的创新平台。该平台跳过传统生病后再被动治疗的思路，致力于在疾病尚处于萌芽阶段时进行拦截与逆转，从而延长人类的健康寿命（Healthspan）。
Viome 自 2017 年推出首款产品以来，业务版图已拓展至全球 100 多个国家和地区。通过直面消费者（Direct-to-Consumer, DTC）等渠道，平台累计汇聚了超过 50 万名用户，并收集了大约 100 万份涵盖粪便、血液和唾液的非侵入式居家生物样本。Banavar 指着大屏幕上的全球用户分布图表示，Viome 最核心的关注点在于建立一个具有极端多样性的大规模人群队列。就在他登台演讲的短短时间内，全球又有数百名新用户加入平台。Banavar 现场向与会者发出邀请："我欢迎大家也成为这场健康运动的一部分"（"I welcome all of you to also become part of the movement here so to speak."）。
- **全球覆盖规模**：来自 100 多个国家和地区，横跨各大洲的 50 多万名真实世界用户。
- **多维度生物样本**：累计回收并测序近 100 万份粪便（肠道微生物）、血液（人体宿主转录组）和唾液（口腔微生物与宿主转录组）样本。
- **人群多样性战略**：摒弃单一地域或特定族群的临床样本局限，持续吸纳全球多样化生活方式与环境背景下的纵向人群数据。
---
### [01:37](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=97s) 押注动态 RNA：跨越人源与微生物群的分子飞轮
在 Viome 创立之初，团队面临着一个决定企业底层技术路线的关键抉择：究竟该押注 DNA、RNA、蛋白质，还是特定代谢物？最终，Viome 团队力排众议选择了 RNA，这一决定直接主导了后续所有技术架构与产品设计的走向。Banavar 解释道，DNA 本质上是一个静态分子，从个体出生、死亡乃至死后数年都几乎一成不变；但在真实活体中，任何特定时间点上通常只有约 5% 的基因被实际表达，这些表达出来的产物正是 RNA。业界普遍认为 RNA 极易降解、极不稳定，Viome 克服了这道工业难题，自主研发出了室温下稳定并保存这一脆弱分子的专利前处理与生信技术，从而能够海量提取人源与微生物源的 RNA 数据。
更重要的是，Viome 没有局限于人类自身的基因表达，而是同步测定人体共生微生物的 RNA，将切入点放在“人源宿主 RNA 与微生物 RNA 的动态互作”上。依托持续汇入的不同表型分子样本，Viome 构建了具备持续学习能力的 AI/ML 预测模型：只要有新的样本表型数据产生，底层模型就会迅速迭代更新，使输出的分子洞见永远代表当前数据的“最佳已知真相”。这也构成了一个独特的转化医学飞轮（Translational Flywheel）——每位通过 DTC 渠道购买检测套件的用户，都在直接为底层科研积蓄数据，进而惠及后续的所有新用户；同时，这套体系最近还催生了临床级的专业诊断检测，向临床医生和医疗机构提供深度的生物分子通路洞见，指导慢病的预防与干预。
```mermaid
flowchart TD
accTitle: Viome Translational R&D Flywheel
accDescr: DTC 用户样本输入、RNA 与微生物测序、AI 学习与临床干预形成的闭环飞轮
subgraph Collection["样本采集与测序"]
A["全球用户居家采样<br/>(粪便 / 血液 / 唾液)"] --> B["RNA 专利常温稳定技术"]
B --> C["宿主与微生物 RNA 测序<br/>(超高通量测序)"]
end
subgraph Engine["AI 计算与模型迭代"]
C --> D["AI / ML 模式识别与特征关联"]
D --> E["持续学习与模型更新<br/>(表型与分子图谱映射)"]
end
subgraph Products["转化医学与干预"]
E --> F["DTC 报告与个性化营养方案"]
E --> G["临床级专业诊断工具"]
F -.->|"驱动新样本与真实世界反馈"| A
G -.->|"临床纵向队列验证"| A
end
```
---
### [04:30](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=270s) 基因表达与微生物组的“共舞”：揭秘环境与疾病的分子桥梁
Viome 的整个技术体系建立在三大基石之上：RNA 技术、微生物组技术与人工智能。谈及为何 RNA 如此关键，Banavar 将其比作连接先天遗传预设与现实生理生活之间的“关键桥梁”。DNA 仅仅记录了先天遗传指令，但个体的饮食构成、睡眠时长、运动活跃度、压力水平以及环境暴露，都在实时决定着基因的开启与关闭。信使 RNA（mRNA）将遗传密码转录并指导下游所有的生物学过程，因此拦截并破译这些正在转录的分子讯息，是解开慢性病发病机制最灵敏的核心抓手。
人体的微生物生态系统则是另一个巨大的未解宝库。虽然过去五六年间该领域的研究才呈现井喷态势，但其生理量级令人惊叹：人体内平均每 1 个属于人类的基因背后，就对应着至少 100 个来自共生微生物的基因；在细胞数量上，人体细胞与微生物细胞的比例也至少达到了 1:1。这些庞大的微生物菌群广泛定植于肠道、口腔乃至皮肤表面。Banavar 形象地指出，所有经口吃下的食物在进入人体前都必须经过肠道微生物这道“筛子”的处理；微生物在代谢食物的过程中会分泌出极其丰富的肠道代谢物，这些代谢物直接触发或调控人体的免疫系统及全身稳态，几乎卷入了所有慢性疾病的病理进程。破解人类宿主基因表达与微生物基因表达之间的这场“分子之舞”，正是 Viome 洞察疾病根源的突破口。
- **基因与细胞数量悬殊**：体内微生物基因量是人类宿主基因量的约 100 倍，微生物细胞与人体细胞比例约为 1:1。
- **代谢筛子效应**：日常饮食经过肠道菌群的发酵分解与二次代谢，生成的代谢物直接影响全身免疫与慢性炎症。
- **环境驱动基因表达**：饮食、作息、运动与精神压力通过 RNA 动态调控通路，弥合静态 DNA 与动态表型之间的盲区。
---
### [07:12](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=432s) PB 级多组学数据与 AI 建模：从血糖反应到生物学年龄预测
作为在 AI 峰会上的演讲，Banavar 重点剖析了 Viome 的多模态海量数据库。得益于横跨 100 多个国家采集的 100 多万份样本，Viome 实验室测序的 RNA 核苷酸序列总量已突破 $10^{15}$（即达到 PB 级，Peta-scale）。将这一极高维度的分子序列与用户详尽的表型数据（涵盖临床确诊病史、用药情况、日常用药、自觉症状、生活方式乃至人口统计学特征）深度对齐后，构建出了高达 $10^{16}$ 维度的庞大生物学数据空间。
利用这些分子特征与临床元数据，Viome 已经构建并验证了超过 30 种重度疾病预测模型，以及数百个尚处于研发阶段的模型。其中数十个成熟模型早已部署在云端生产环境，支撑着普通用户移动端 App 背后的实时分析系统。
Banavar 在现场展示了几个极具代表性的 AI 预测维度：
1. **个体化餐后血糖反应（Glycemic Response）预测**：针对食物本体库中的数百种食物，模型可精准预测特定用户进食后的血糖波动幅度（是平缓稳定还是剧烈飙升），帮助用户精准避开导致个人血糖激增的食物。该算法完全建立在严谨的临床对照研究之上。
2. **分子生物学年龄（Biological Age）评估**：系统整合粪便、血液与唾液样本的整体分子转录特征，将其映射到数十万大规模人群的分子基准线中。例如，一位实际年龄 32 岁的用户，其分子档案若与 38 岁人群高度吻合，系统便会判定其生物学年龄已达 38 岁，为其过早衰老提供警示与生活方式干预依据。
3. **全身通路与诊断模型**：模型还能准确反推用户的睡眠质量、质子泵抑制剂（PPIs）和抗生素的使用史，乃至用户出生时是顺产还是剖宫产。在复杂通路层面，算法打通了肠道微生物组、血液转录组与唾液宏转录组（Metatranscriptome），涵盖认知功能通路、牙科与口腔健康通路、肠道炎性通路等。演讲最后，Banavar 进一步透露，Viome 在高危病种上已延伸至肿瘤筛查，成功构建了针对口腔癌与咽喉癌（Oral and Throat Cancer）的分子诊断模型。
- **数据吞吐量**：实验室测序逾 $10^{15}$ 个 RNA 核苷酸（PB 级），结合表型后累积达 $10^{16}$ 个生物学数据点。
- **落地模型资产**：30+ 种疾病预测模型落地，数百个 R&D 模型在研，数十个核心模型在线服务。
- **跨组学通路覆盖**：融合肠道菌群转录组、血液宿主转录组与唾液宏转录组，实现从生活习惯回溯到恶性肿瘤（如口腔咽喉癌）筛查的全链条覆盖。
### [11:15](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=675s) AI 模型体系与分子通路建模：从生物标志物到基础模型
Viome 在过去八年间积累了庞大的非侵入式样本库，能够通过唾液样本精准识别口腔癌和咽喉癌的特异性分子特征。在此基础上，团队构建了一套先进的 AI 与机器学习工具链。这套体系不仅包含常规的生物信息学数据分析，还深入整合了分子通路模型与智能推荐引擎。针对日常认知中“某些食物普遍健康”的误区，推荐引擎能够根据个体的微观通路状态进行动态矫正。例如，菠菜虽常被视作健康蔬菜，但若个体的肠道草酸盐代谢通路处于受损或低效状态，摄入过量菠菜会导致草酸盐超负荷积聚，进而显著增加罹患肾结石的风险。
除了通路建模，Viome 还将大语言模型（LLM）引入用户体验端，把复杂的生物学洞察转化为通俗易懂的个性化健康报告。在诊断算法层面，团队正在研发分子基础模型（Molecular Foundation Models），该模型展现出了作为通用诊断分类器的巨大潜力。
```mermaid
flowchart TD
accTitle: Viome AI 推荐引擎决策流
subgraph InputData["多组学多组织数据输入"]
A["唾液/血液/粪便样本 (RNA)"] --> B["生物信息学与分子通路模型"]
end
subgraph Evaluation["双重生理评估机制"]
B --> C["宏观营养评估<br/>(血糖反应预测)"]
B --> D["微观通路评估<br/>(草酸代谢/TMA/气体生成)"]
end
subgraph Decision["个性化分类输出"]
C & D --> E{"综合决策引擎"}
E -->|"通路通畅 & 血糖平稳"| F["推荐享用 (Enjoy)"]
E -->|"通路受阻 (如草酸超负荷)"| G["降级避开 (Avoid/Minimize)"]
end
```
- **先进 AI 架构组分**：生物信息学数据处理管线、分子通路模拟器、个性化饮食/补充剂推荐引擎、用于洞察交付的 LLM 交互层，以及作为诊断分类器的分子基础模型。
- **代表性生化机制推断**：草酸代谢通路（Oxalate Metabolism Pathway）异常时，高草酸饮食（如大量菠菜）直接关联肾结石风险。
---
### [12:48](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=768s) 多组学健康分区评分与 20+ 种疾病风险预测
基于 RNA 与微生物组数据，Viome 构建了一套多维度的健康评分体系，并在移动端 App 中以直观的“健康分区（Health Zones）”呈现，涵盖肠道、口腔、心血管、肾脏、心脏以及心理健康等核心生理领域。目前，该系统能够依据多通路测量结果，同时评估并向用户展示约 20 种常见慢性疾病的潜在风险。这种基于粪便、血液和唾液的多组学全景图谱，能够帮助预防医学医生针对不同患者的特定关注点进行精准筛查。
在健康分区之下，系统会将生理状态进一步拆解至精细的分子通路层级，并提供生物学年龄（Biological Age）与全身整体炎症负荷评分。例如，App 能直接揭示特定分子通路与牙周病、干燥综合征（Sjögren's syndrome）或多动症（ADHD）的关联机制。在一个具体的心脏病风险案例中，检测发现受试者的低密度脂蛋白胆固醇（LDL Cholesterol）通路、甲烷气体生成通路以及三甲胺（TMA）生成通路同时处于欠佳的“警示区间（Attention Zone）”，系统由此精确锁定了导致其心血管发病风险升高的分子根源。
- **多组织生物样本**：粪便、血液与唾液三位一体的数据采集。
- **健康分区与长寿指标**：肠道、口腔、心脏、肾脏、心理等健康分区，结合全身综合炎症指数与生物学年龄。
- **疾病关联示例**：心脏病风险直接溯源至三甲胺（TMA）生成异常、甲烷代谢紊乱与 LDL 胆固醇通路异常。
---
### [15:19](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=919s) 精准营养推荐引擎与食物本体库
对于便秘等常见临床症状，传统的泛化干预往往收效甚微，因为肠易激综合征（IBS）便秘型可能由截然不同的底层生化异常引起——包括硫化物气体超标、甲烷气体滞留或次级胆汁酸代谢失衡。Viome 的推荐引擎会精准锁定具体的异常通路，从而定制针对性的营养解决方案，而非提供一刀切的通用饮食建议。
该系统依赖一套庞大的食物本体数据库（Food Ontology）。推荐引擎首先分析食物的宏量营养素以预测个体餐后血糖反应，将食物初步归入“推荐享用（Enjoy）”或“尽量减少（Minimize）”类别；随后，算法进一步扫描微量营养素及其对下游分子通路的影响。如果个体的特定代谢机制（如草酸清除通路）存在缺陷，即使某种食物具有优异的低升糖指数，也会被算法自动降级为“避免食用（Avoid）”。App 会为每种食物附上详尽的科学依据，解释其推荐或禁忌的分子原因；对于不便通过日常膳食调整的用户，平台还会提供高浓度的个性化定制补充剂，以靶向修复受损通路。
- **复杂症状的分型溯源**：针对便秘症状，精准区隔硫化物气体、甲烷气体及次级胆汁酸通路。
- **分级过滤算法**：宏量营养素血糖预测（初筛） $\rightarrow$ 微量营养素通路影响评分（微调与降级）。
- **个性化干预形式**：精准食谱指导与定制化浓缩成分营养补充剂。
---
### [17:06](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1026s) 临床随机对照试验验证与企业员工健康落地成效
为了验证这套“数据分析-分子通路-精准干预”闭环的实际疗效，Viome 开展了严格的随机双盲安慰剂对照试验（RCT）。在针对肠易激综合征（IBS）便秘患者的临床试验中，接受个性化干预方案的受试组达到了 64% 的症状完全缓解率，而安慰剂对照组的缓解率仅为 10% 左右。类似的临床显著性与统计学显著差异在焦虑症和糖尿病的干预试验中同样得到了充分证实。
除了严格的医学临床试验，该方案还在大型企业员工健康管理项目中进行了实证落地。在一项为期六个月的企业全员试点中，采用 Viome 分析与精准干预方案的员工群体表现出了显著的生理与职场表现改善：58% 的参与者成功减轻了体重，100% 的受试者反馈精力水平显著提升。更为关键的是，肠道健康的改善直接提升了员工在工作场景中的专注度与生产力，为雇主与员工双方创造了切实双赢的价值。
- **IBS 临床试验数据**：干预组便秘症状消除率达 64%，安慰剂对照组约为 10%。
- **企业健康试点成果**：为期 6 个月的干预使 58% 参与者体重下降，100% 报告体能与精力改善。
- **综合效益溢出**：生活质量改善同步转化为更高的职场工作专注度与工作效率。
---
### [18:45](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1125s) 商业化生态：面向消费者、企业与医疗机构的精准健康体系
在创立八年的发展历程中，Viome 已经构建起覆盖多维应用场景的产品矩阵，并持续迭代其底层算法与检测技术。该体系目前已实现多轨并行：消费者可以直接订购 C 端检测服务与个性化干预方案；企业雇主能够将整套解决方案无缝集成至员工福利与健康管理计划中。
面向临床医生与医疗专业机构，Viome 专门推出了临床级的“Precision Health Pro”检测服务。该版本为预防医学与功能医学医生提供了更高精度的患者分层数据、更深入的分子通路指导以及每一项分析背后的详尽科学循证支持。演讲最后，CTO Guru Banavar 呼吁现场的雇主代表与医疗从业者积极尝试这一工具，共同推动基于分子通路的精准预防医学普及。
- **三大业务形态**：直接面向消费者（DTC）、企业员工健康福利方案（Employer Wellness）、临床级专业检测（Precision Health Pro）。
- **专业端赋能**：为临床医生提供更深度的风险分层、临床指导建议与循证科学支持。
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=4s) I welcome all of you to also become part of the movement here so to speak. — Guru Banavar
> [01:37](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=97s) DNA 本质上是一个静态分子，从个体出生、死亡乃至死后数年都几乎一成不变；但在真实活体中，任何特定时间点上通常只有约 5% 的基因被实际表达，这些表达出来的产物正是 RNA。 — Guru Banavar
> [04:30](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=270s) 人体内平均每 1 个属于人类的基因背后，就对应着至少 100 个来自共生微生物的基因；在细胞数量上，人体细胞与微生物细胞的比例也至少达到了 1:1。 — Guru Banavar
> [04:30](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=270s) 所有经口吃下的食物在进入人体前都必须经过肠道微生物这道“筛子”的处理；微生物在代谢食物的过程中会分泌出极其丰富的肠道代谢物，这些代谢物直接触发或调控人体的免疫系统及全身稳态。 — Guru Banavar
> [11:15](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=675s) 菠菜虽常被视作健康蔬菜，但若个体的肠道草酸盐代谢通路处于受损或低效状态，摄入过量菠菜会导致草酸盐超负荷积聚，进而显著增加罹患肾结石的风险。 — Guru Banavar
> [15:19](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=919s) 针对便秘等常见临床症状，传统的泛化干预往往收效甚微，因为肠易激综合征（IBS）便秘型可能由截然不同的底层生化异常引起——包括硫化物气体超标、甲烷气体滞留或次级胆汁酸代谢失衡。 — Guru Banavar
> [15:19](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=919s) 如果个体的特定代谢机制（如草酸清除通路）存在缺陷，即使某种食物具有优异的低升糖指数，也会被算法自动降级为“避免食用（Avoid）”。 — Guru Banavar
> [17:06](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1026s) 在针对肠易激综合征（IBS）便秘患者的临床试验中，接受个性化干预方案的受试组达到了 64% 的症状完全缓解率，而安慰剂对照组的缓解率仅为 10% 左右。 — Guru Banavar
> [17:06](https://www.youtube.com/watch?v=J7u6DCtOYnU&t=1026s) 在一项为期六个月的企业全员试点中，采用 Viome 分析与精准干预方案的员工群体表现出了显著的生理与职场表现改善：58% 的参与者成功减轻了体重，100% 的受试者反馈精力水平显著提升。 — Guru Banavar
## Mentioned
Guru Banavar, Eric, Viome Life Sciences, Viome, Ai4, Precision Health Pro, RNA, mRNA, DNA, 微生物组 (Microbiome), 大语言模型 (LLM), 分子基础模型 (Molecular Foundation Models), 食物本体数据库 (Food Ontology), 肠易激综合征 (IBS), 干燥综合征 (Sjögren's syndrome), 注意力缺陷多动障碍 (ADHD), 口腔癌与咽喉癌 (Oral and Throat Cancer), 三甲胺 (TMA), 低密度脂蛋白胆固醇 (LDL Cholesterol), 质子泵抑制剂 (PPIs)
## Video description
Guru Banavar, Founding CTO, Viome Life Sciences
Abstracts
Chronic disease is a major epidemic in the 21st century, with more than 50% of the world population suffering some form of chronic disease leading to lowered quality of life and shorter lives. The vast majority of chronic diseases are due to environmental and lifestyle factors having a major impact on our molecular pathways. At Viome, our mission is to detect disruptions in those pathways as early as possible, and provide health insights and interventions that are precisely tailored to individuals to maintain a healthy state. To this end, we have developed a platform that has extracted peta-scale gene expression data from the microbial and human host cells from more than a million non-invasive samples collected at home from individuals in over 100 countries. We have analyzed this high-dimensional and longitudinal data using AI and ML techniques, discovered a range of host-microbiome biomarkers associated with a range of chronic diseases, derived health insights from these biomarkers, and delivered personalized interventions through a suite of consumer and professional products. In this talk, I will present how this approach can lead to a new era of molecular health for the global population.
[Source](https://www.youtube.com/watch?v=J7u6DCtOYnU)
Source quotes
[0:04] Guru Banavar: I welcome all of you to also become part of the movement here so to speak.
[1:37] Guru Banavar: DNA 本质上是一个静态分子，从个体出生、死亡乃至死后数年都几乎一成不变；但在真实活体中，任何特定时间点上通常只有约 5% 的基因被实际表达，这些表达出来的产物正是 RNA。
[4:30] Guru Banavar: 人体内平均每 1 个属于人类的基因背后，就对应着至少 100 个来自共生微生物的基因；在细胞数量上，人体细胞与微生物细胞的比例也至少达到了 1:1。
[4:30] Guru Banavar: 所有经口吃下的食物在进入人体前都必须经过肠道微生物这道“筛子”的处理；微生物在代谢食物的过程中会分泌出极其丰富的肠道代谢物，这些代谢物直接触发或调控人体的免疫系统及全身稳态。
[11:15] Guru Banavar: 菠菜虽常被视作健康蔬菜，但若个体的肠道草酸盐代谢通路处于受损或低效状态，摄入过量菠菜会导致草酸盐超负荷积聚，进而显著增加罹患肾结石的风险。
[15:19] Guru Banavar: 针对便秘等常见临床症状，传统的泛化干预往往收效甚微，因为肠易激综合征（IBS）便秘型可能由截然不同的底层生化异常引起——包括硫化物气体超标、甲烷气体滞留或次级胆汁酸代谢失衡。
[15:19] Guru Banavar: 如果个体的特定代谢机制（如草酸清除通路）存在缺陷，即使某种食物具有优异的低升糖指数，也会被算法自动降级为“避免食用（Avoid）”。
[17:06] Guru Banavar: 在针对肠易激综合征（IBS）便秘患者的临床试验中，接受个性化干预方案的受试组达到了 64% 的症状完全缓解率，而安慰剂对照组的缓解率仅为 10% 左右。
[17:06] Guru Banavar: 在一项为期六个月的企业全员试点中，采用 Viome 分析与精准干预方案的员工群体表现出了显著的生理与职场表现改善：58% 的参与者成功减轻了体重，100% 的受试者反馈精力水平显著提升。
Source quotes
Guru Banavar: Chronic disease is a major epidemic in the 21st century, with more than 50% of the world population suffering some form of chronic disease leading to lowered quality of life and shorter lives.
Guru Banavar: The vast majority of chronic diseases are due to environmental and lifestyle factors having a major impact on our molecular pathways.
Guru Banavar: At Viome, our mission is to detect disruptions in those pathways as early as possible, and provide health insights and interventions that are precisely tailored to individuals to maintain a healthy state.
Guru Banavar: To this end, we have developed a platform that has extracted peta-scale gene expression data from the microbial and human host cells from more than a million non-invasive samples collected at home from individuals in over 100 countries.
Guru Banavar: We have analyzed this high-dimensional and longitudinal data using AI and ML techniques, discovered a range of host-microbiome biomarkers associated with a range of chronic diseases, derived health insights from these biomarkers, and delivered personalized interventions through a suite of consumer and professional products.
Guru Banavar: In this talk, I will present how this approach can lead to a new era of molecular health for the global population.
[0:04] Guru Banavar: While DNA is static and remains largely unchanged from birth until long after death, RNA is dynamic—only about 5% of genes are actively expressed at any given point in time.
[4:31] Guru Banavar: While there is roughly one microbial cell for every human cell in the body... for every single human gene, there are approximately 100 microbial genes.
Original transcript
[0:04] Hi everybody. Uh thank you very much for
[0:06] uh having me over here. Um so I would
[0:09] like to look at a very different
[0:11] spectrum of the healthcare industry so
[0:15] to speak. It's uh you just heard from
[0:17] Eric the uh drug development side. I
[0:20] would like to talk to you about a few
[0:22] products and a platform that's already
[0:24] in market and it's helping people
[0:26] prevent
[0:28] chronic disease and uh hopefully improve
[0:32] lifespan. I'm going to discuss how we're
[0:34] doing that through this platform. Um let
[0:37] me make sure that this works.
[0:39] Yeah. So by way of introduction um Viome
[0:43] has been in the market since 2017.
[0:47] We've had products in the market. We
[0:48] have customers in more than 100
[0:51] countries around the world. We have a
[0:53] half a million people who have sent us
[0:55] approximately a million samples, stool,
[0:58] blood and saliva. And we have focused
[1:01] our attention on this very large scale
[1:04] human population which you can see uh
[1:06] over here.
[1:10] You can see that human population has
[1:12] been our our focus. So we want to
[1:15] collect as much data as possible from as
[1:17] diverse a population as possible from
[1:20] around the world and we have people
[1:23] joining our platform through our
[1:25] products every single day. Since the
[1:27] time I got to this conference there's
[1:30] probably several hundred more people. So
[1:32] I welcome all of you to also become part
[1:34] of the movement here so to speak. Um we
[1:37] have focused on RNA as our molecule of
[1:41] choice and we had to make a tough
[1:43] decision when we started biome um to
[1:46] whether to focus on DNA, RNA, proteins
[1:50] or certain sets of metabolites if
[1:52] necessary and so on. We decided that we
[1:54] would go with RNA and that gave rise to
[1:58] a whole set of downstream decisions that
[2:02] we had to make. um we in fact not only
[2:05] looked at human RNA, we also looked at
[2:07] microbial RNA and um we decided that the
[2:11] interaction of the human and the
[2:14] microbial RNA is the place to focus for
[2:16] understanding chronic disease. As you
[2:18] know DNA is a static molecule. Um it is
[2:22] more or less unchanged from the time
[2:24] you're born till the time you die and
[2:26] even after many many years after you
[2:27] die. But what is changing in your body
[2:30] constantly is RNA.
[2:32] About 5% of your genes are expressed at
[2:36] any given point in time and that becomes
[2:38] the RNA. Now a lot of people think that
[2:40] RNA is an unstable molecule. It is true.
[2:43] But we have come up with techniques for
[2:44] stabilizing this unstable molecule and
[2:47] collecting tons of data from both the
[2:51] human side and the microbial side. And
[2:53] we have analyzed all of this data using
[2:55] a whole collection of AI and ML modeling
[2:59] and analysis tools from um of the point
[3:03] of view of understanding the phenotypes
[3:06] associated with various molecular
[3:08] patterns. And also we have this model of
[3:11] continuous learning as in when we get
[3:14] more samples of a particular phenotype
[3:16] we update our models as quickly as
[3:19] possible so that we are all always
[3:22] representing what is the best known
[3:24] truth according to our data and so we
[3:27] have created a number of products some
[3:29] of our products are direct to consumer
[3:31] products and I'll show you one or two
[3:32] examples of that that actually creates a
[3:35] um I would call it a flywheel effect
[3:37] because every customer who comes into
[3:40] Viome through our direct to consumer
[3:41] channel is contributing to the science
[3:44] and the progress for all other customers
[3:46] who come after them. So we've created um
[3:49] very recently also some professional
[3:51] clinical grade tests that is also
[3:53] becoming even more insightful in terms
[3:56] of the pathways that we analyze and the
[3:58] insights we give to providers who can
[4:01] then help patients prevent and reverse
[4:03] chronic disease. And so that innovative
[4:06] cycle of R&D has become a translational
[4:08] flywheel because we believe in being in
[4:11] the market, understanding what's going
[4:12] on in the field, looking at longitudinal
[4:14] samples. And to do all of this stuff,
[4:17] we've relied on three key technologies.
[4:19] RNA technologies,
[4:22] microbiome technologies, and AI.
[4:25] So I would like to walk you through each
[4:27] of these three technologies and how we
[4:28] use it and how we've applied it. You
[4:31] probably most of you already know that
[4:34] RNA is a very important molecule in the
[4:37] body. It is the bridge between what was
[4:40] genetically predetermined in your DNA
[4:42] and what actually happens in real life
[4:44] depending on what you're doing, what you
[4:46] eat, how you sleep, how you um how much
[4:50] how active you are and how much stress
[4:53] you have and everything else decides
[4:55] what kind of gene expression happens. So
[4:58] it is a messenger RNA that transmits the
[5:04] genetic code into a lot of downstream
[5:07] processes. And we think that getting a
[5:10] sort of an interception of that message
[5:13] and understanding what is being trans um
[5:17] transcribed from your DNA is the
[5:19] critical point for understanding chronic
[5:21] disease. Now the human microbiome is a
[5:24] fascinating field. Um many of you have
[5:26] probably already heard about this. Maybe
[5:27] some of you have already worked in this,
[5:29] but for every gene in your human body,
[5:34] there are probably a 100 genes from a
[5:38] microbial ecosystem in your body. Now,
[5:42] in terms of the number of cells, for
[5:43] every human cell, there's at least one
[5:45] microbial cell in your body. A lot of it
[5:47] is focused in your mic in your gut. A
[5:50] lot of it is also in your mouth, and
[5:52] it's in many other places on your body,
[5:54] on your skin, and everywhere else in
[5:55] your body. But understanding the role of
[5:58] the microbiome has not been a very
[6:01] active field of research until the last
[6:03] maybe five or or so years and biome has
[6:08] focused on the role of the microbiome
[6:10] because the microbiome has been
[6:11] implicated in many if not all chronic
[6:15] diseases. There is a role that the
[6:16] microbiome has to play triggers a lot of
[6:20] immune function. It triggers a lot of
[6:22] other important functions for gut
[6:25] metabolites. You know, everything you
[6:27] eat goes through a sort of a civ in your
[6:30] gut. Your microbiome is in fact
[6:32] processing your food and then generating
[6:36] a number of gut metabolites that impact
[6:38] your health and wellness tremendously.
[6:41] And we have a lot of data now to show
[6:43] what the impact of your gut metabolites
[6:45] are. So we have focused on understanding
[6:49] the interaction between your human gene
[6:53] expression in your human cells as well
[6:56] as the microbial gene expression and
[7:00] what the sort of the interaction and the
[7:02] dance between those two things are and
[7:04] that has given rise to a number of uh
[7:07] insights that that I will be happy to go
[7:09] through. Now um since we're on an AI
[7:13] conference I have to first talk about
[7:14] the data right we have collected more
[7:17] than a million samples now from
[7:19] individuals in more than 100 countries
[7:22] around the world because these
[7:23] individuals have purchased the Viome
[7:27] direct to consumer test and uh sent in
[7:31] their samples from all over the world
[7:32] from all the continents around the
[7:34] world. And so that data this is real
[7:37] world data that data has resulted in
[7:40] more than 10 to the 15 so that's pascale
[7:44] RNA nucleotides that have been now
[7:47] sequenced in our lab at bio and that has
[7:51] also been combined with a tremendous
[7:54] amount of phenotype data that we collect
[7:55] from the same individuals and so the
[7:57] total number of biological data points
[7:59] is in the in the order of 10 to the 16
[8:01] and that has resulted in a number of
[8:05] what I would call important features
[8:06] like the genes, the organisms from
[8:10] various samples. We've we've uh
[8:12] extracted all of this insight from the
[8:15] data. We have also looked at all of the
[8:17] phenotype metadata such as the symptoms,
[8:20] the diseases, the drugs people take, the
[8:22] medications, the lifestyle demographics.
[8:24] All of that data from all this huge
[8:27] collection of people who've come into
[8:29] Viome has been collected and has been
[8:32] correlated with the molecular features.
[8:35] So when you combine the molecular
[8:37] features with the phenotype metadata,
[8:39] you can create a number of models
[8:42] depending on what you're looking for.
[8:44] And we have been so far able to to build
[8:47] at least 30 important disease models and
[8:50] we have literally hundreds of other R&D
[8:52] models that that we are currently still
[8:55] building. Many of these models are
[8:56] already in production. So if you are a
[8:58] customer today, if you come in and
[9:00] purchase a Viome kit and you get your
[9:02] results in an app,
[9:05] there are dozens of AI models that are
[9:08] working behind the scenes to generate
[9:10] that data for you today. So I want to
[9:14] show you some of the models that we've
[9:16] built. Uh for example, we can predict
[9:19] your glycemic response.
[9:22] Um in other words, when you eat any
[9:24] food, what is your blood sugar spike
[9:28] likely to be? Is it flat? Is it is it
[9:31] extreme? Is it something that you should
[9:33] avoid or minimize eating the food for
[9:36] because your blood sugar is spiking? And
[9:38] we did an entire clinical study just to
[9:40] understand the glycemic response for any
[9:42] food in our food ontology which is
[9:44] hundreds and hundreds of foods. We can
[9:46] also predict your what we call the
[9:48] biological age. Meaning if you send in
[9:52] your stool sample, your blood sample and
[9:54] your saliva sample, we can tell you what
[9:56] is your molecular pattern or molecular
[9:59] profile looking like in terms of the
[10:01] closest age that we've seen in our
[10:04] entire large cohort of hundreds of
[10:06] thousands of samples. So we can tell
[10:08] you, for example, if you're 32 years old
[10:10] and we tell you that your your
[10:12] biological age is more like 38, what it
[10:15] means is that your molecular profile
[10:17] looks like a cohort of 38 year old
[10:20] people, not 32 year old people. And
[10:22] that's a cause for concern. And that can
[10:24] be something that you can use for
[10:26] understanding what to do in the future.
[10:27] And then I don't have time to walk
[10:29] through all of these models, but we can
[10:31] predict your sleep quality, your use of
[10:33] PPIs and antibiotics, whether you were
[10:35] born with a C-section or not. We can
[10:38] figure out what kinds of pathways, all
[10:41] of these blue ones are pathways, whether
[10:43] it is your cognitive function pathways,
[10:45] dental pathways, gut inflammatory
[10:48] pathways. We we predict all of these
[10:50] pathways using your molecular data as a
[10:54] starting point and connecting the dots
[10:56] between your gut microbiome, your blood
[10:58] transcriptto and your salivary metat
[11:01] transanscriptto. And finally, these um
[11:04] red uh models that you see out here are
[11:06] also diagnostic models that we've built
[11:09] where we've gone all the way to cancers.
[11:11] For example, oral and throat cancer. We
[11:13] have a cohort of oral and throat cancer
[11:15] patients and a cohort of healthy people
[11:18] and we can tell what is the signature of
[11:20] oral and throat cancer and we can detect
[11:22] that from a salivary sample. Right? So
[11:24] this is the set of samples that we've
[11:26] built over the last 8 years or so and
[11:29] we've combined all of that into a number
[11:32] of I would just call them advanced AI
[11:35] models and methods. You know, we have
[11:37] bioinformatics, which you can imagine is
[11:39] is a pretty standard part of a a tool
[11:41] set like this. We have these molecular
[11:43] pathway models. We have a recommendation
[11:45] engine that takes into account your
[11:47] pathways and tells you what to do, what
[11:49] not to do. Such as you might think that,
[11:52] you know, eating certain types of foods
[11:54] is is always healthy. It's it's
[11:56] universally healthy. It turns out that's
[11:59] not the case. There's many kinds of
[12:01] foods that may that are generally known
[12:03] to be healthy. For example, spinach is
[12:06] known to have is known to be very
[12:07] healthy, but it can actually exacerbate
[12:09] your oxalate metabolism pathways in your
[12:12] gut and that may result if your oxalate
[12:15] metabolism pathway is not doing very
[12:16] well and you eat a tremendous amount of
[12:18] spinach, it can create an oxalate
[12:20] overload and it may create kidney
[12:22] stones. Those are the kinds of um
[12:25] inferences we can make based on this
[12:27] recommendation engine. We also use LLMs
[12:30] for user experience for people to
[12:32] understand what their health insights
[12:34] are. If you can see our see our app, we
[12:36] do a number of biomarker discoveries.
[12:38] And finally, we also have some molecular
[12:40] foundation models that are promising in
[12:43] terms of its ability to be a diagnostic
[12:45] classifier as well. Um, so once we put
[12:49] all of that together, the RNA and
[12:50] microbiome data results in a number of
[12:54] these health scores. And I'll show you a
[12:57] few examples of the health scores. And
[12:58] then we can correlate that to disease
[13:01] risks. We currently have something like
[13:04] 20 diseases that we can show you the
[13:06] risks for based on all of the pathways
[13:09] that we measure through our health
[13:10] scores. And that shows up in your app as
[13:14] soon as you do your test. And in in
[13:16] fact, this is a type of um picture that
[13:19] a provider a healthcare provider would
[13:22] be very interested in seeing. I was just
[13:24] talking to uh my friend over here Tao
[13:26] and he he was saying that you know he'd
[13:28] be interested in certain diseases but
[13:30] every doctor that we go to who is
[13:32] interested in preventive medicine could
[13:33] potentially be interested in a different
[13:35] set of diseases and we give you your
[13:37] risk levels for all of these diseases
[13:40] using your stool blood and saliva data.
[13:43] We also provide all of these inflam
[13:45] inflammation and so forth. Of course as
[13:47] we get more data we publish these
[13:49] results. I've shown a few of the
[13:51] publications down here, but there's many
[13:53] other publications that we consistently
[13:55] produce. This is how the app looks like.
[13:57] If you get your results today, you get
[13:59] to see your what we call your health
[14:02] zones, you know, which are basically
[14:03] your gut zone, your oral, your
[14:06] cardiovascular, kidney, heart zone, your
[14:10] mental health zone, and so forth. And
[14:12] for each of those zones, we break it
[14:15] down into many different pathways. You
[14:18] also get your longevity scores like your
[14:19] inflammation. What is the total amount
[14:21] of inflammation we see across your
[14:22] entire body? You can see your biological
[14:25] age as I said before. And you know when
[14:27] you break down the pathways you can see
[14:29] down here several pathways that are have
[14:32] been connected to gum disease or shogun
[14:36] syndrome or ADHD. These are all the
[14:38] disease risks that we show you in your
[14:41] app when you when you get your results
[14:43] back. um and each of these disease areas
[14:47] has been shown to be associated with a
[14:50] particular pathway. In this case, this
[14:53] individual has a risk of heart disease.
[14:56] And you can see here that there are
[14:58] three. There's an LDL cholesterol
[14:59] pathway, there's a methane gas
[15:00] production pathway, and there's a TMA
[15:02] production pathway. All three of which
[15:05] are suboptimal. They are in the
[15:07] attention zone. And as a result of that
[15:09] you can see why the TMA production
[15:12] pathway was implicated in the risk for
[15:16] heart disease. Now
[15:19] for a given symptom like constipation
[15:22] for example you could have a number of
[15:24] different pathways that are responsible
[15:26] for it. It's not just one thing when you
[15:28] look at IBS constipation. So for example
[15:30] in this case we can see sulfide gas,
[15:32] methane gas, secondary bile acids etc.
[15:34] All of these are potentially problematic
[15:37] if you had constipation as an issue. We
[15:39] can see each one of those pathways and
[15:41] give you the appropriate nutritional
[15:44] program to address that particular
[15:46] pathway and not a generic thing for
[15:47] addressing um constipation.
[15:50] How do we do that? We have a huge food
[15:55] database or a food ontology from which
[15:57] we can see the macronutrients and first
[16:00] predict your glycemic response for each
[16:02] of those foods.
[16:05] Based on that, we can then see whether a
[16:07] food is in in your enjoy food bucket or
[16:10] your minimize food bucket depending on
[16:11] whether it spikes or does not spike. And
[16:14] furthermore, we can see all the
[16:16] micronutrients and look for all of the
[16:18] pathways that I was just talking about.
[16:20] Like for example in an oxalate
[16:21] elimination pathway if your oxalate
[16:23] elimination pathway is not doing great
[16:25] then even a food that has a low glycemic
[16:28] response meaning it's in the enjoy
[16:30] bucket can get demoted to being an avoid
[16:35] food because you don't want to uh
[16:38] process all the oxalates coming from
[16:39] spinach as an example. Right? This is
[16:41] how our recommendation engine works. And
[16:44] you can see in your app for every single
[16:46] food whether it makes sense to eat
[16:49] enough of a lot of that food or limit
[16:51] that food and so on. And it explains the
[16:53] reasons why a given food should be
[16:56] avoided, should be minimized, should be
[16:59] enjoyed etc etc. So these are all the
[17:01] kinds of scientifically based um results
[17:05] that we show. And in fact, if you don't
[17:07] want to um just look at foods, you can
[17:10] also get supplements that could be
[17:13] concentrated forms of that same food or
[17:15] the ingredients from that food and that
[17:17] can potentially improve your pathways as
[17:19] well. We've done randomized placeboc
[17:22] control trials with this entire system
[17:25] and checked whether the control arm,
[17:27] which is placeboc controlled, and the
[17:30] intervention arm, is there a significant
[17:32] difference? And we've seen for IBS
[17:34] there's a 64% uh chance of getting a uh
[17:39] person who had constipation into a state
[17:42] of no symptoms versus I think it's 10%
[17:45] in the placebo arm and so on with
[17:48] anxiety as well with diabetes as well
[17:50] we've we've shown that there's a
[17:51] significant clinical as well as a
[17:53] statistical difference between the
[17:55] control arm and the intervention arm and
[17:57] finally I just want to say that we've
[17:59] done largecale studies within
[18:01] organizations conditions when an
[18:03] organization which has a wellness
[18:05] program for example wants to implement a
[18:09] program with the whole Viome analysis
[18:12] and the Viome interventions we've seen
[18:15] weight reduction 58% of the people that
[18:17] we did it this particular pilot in lost
[18:20] weight 100% of the people reported
[18:22] improved energies after a six-month
[18:25] program programmatic uh implementation
[18:28] of of biomes um analysis now Here I also
[18:32] want to say that gut health and quality
[18:34] of life by the way it's not just quality
[18:37] of life in their personal life but also
[18:38] their productivity and their focus
[18:40] improved. So it was a benefit for the
[18:42] employer as well as for the employees.
[18:44] So just to summarize and wrap up um Viom
[18:49] has been around for now eight years and
[18:53] has established a set of products which
[18:55] have continuously improved and will
[18:57] continue to improve going forward. And
[19:00] this is available for I don't know what
[19:02] that is uh but this is available for
[19:05] consumers directly. So each one of you
[19:07] can purchase this directly or if you're
[19:09] an employer you might consider taking
[19:12] this to your entire organization. If
[19:13] you're a provider we have a clinical
[19:15] grade test called the precision health
[19:17] pro test that gives you additional
[19:19] information additional stratification
[19:21] additional guidance additional science
[19:24] behind every single analysis and insight
[19:26] that's been that's been presented here.
[19:28] And so with that, I'm going to, you
[19:30] know, stop by saying, uh, please try out
[19:33] this test. If you're interested in
[19:35] implementing it in your organization,
[19:37] whether you're an employer or you are a
[19:40] provider, please contact me. I put in a
[19:42] um a QR code over there. You can
[19:44] actually take a picture of that and you
[19:45] will go to a little form where you can
[19:47] enter your information and we'll talk to
[19:49] you about exactly how to implement this
[19:51] in your organization. So, thank you very
[19:52] much. Appreciate it.
[19:55] [Applause]
