Apps and architecture for human superpowers and longer life
We are trying to answer the urgent question, “how will we defend a human from cancer?” My starting point was a more abstract question, “how will our apps and AI advisers get to know us?” Digital advisers get better as they get access to more data. In order for humans to use the amazing resources available on the Internet and in AI labs, we need a way to jack in. But humans don’t have APIs. And soon, the data about a human will be too big and too private to move around the Internet, or carry on a phone. When I wrote up a proposal to solve this problem, my reviewers laughed. It seemed remote and even silly to them.
Then a friend called me to say “I’ve been diagnosed with a dangerous form of cancer.” He was carrying around a hard drive with a terabyte of genetic sequencing data, shopping for something called a neoantigen vaccine. He wanted to know if I could streamline the process. That was the beginning of the HumanDB project.
Apps can help treat cancer, if they know enough about you. We are approaching a point where machines can read the DNA and RNA of a cancer, understand how it works and what it looks like to your immune system, and figure out how to treat it. They will also give you superpowers. They will shop for you, invest for you, move you around, and identify career opportunities. They can give you memory superpowers, remembering everything you see or hear. If you believe Kurzweil, in the future we may even merge brains with our advancing digital twins.
Trending AI Articles:
HumanDB — one for each human
To do this, you will need a computer that represents you, and runs apps for you. It’s like your mobile phone, except that it lives in the cloud, where you can’t forget it or drop it in the toilet.
We will gather your data in one place, and then bring apps to the data.
This structure has a lot of benefits that we don’t get from the software at your local hospital.
Users can install a HumanDB anywhere. They can use their own computers, or our SaaS service, or AWS in the US, or Alibaba in China. Local installation eliminates the barriers created by laws that prevent personal data from traveling across national boundaries. It creates a channel for distributing medical knowledge globally.
Eliminates barriers caused by data privacy laws
Data privacy laws make it difficult to use health data. Organizations that have information are reluctant to send it out, because they have a legal responsibility to keep it private. Personal databases eliminate this problem. It’s easier to get the data, because sending data to its human owner does not create a liability. It’s easier to use the data, because humans have the right to use and share their own information. Local apps eliminate a lot of data privacy problems by sending less data to vendors. Users that care about privacy can select hosts that have good data protection and encryption.
Unlimited data size
We can make our apps and analysts smarter by giving them large amounts of data. We do not need to send the data across a network for each use. We can accumulate large databases because the data only needs to move once, when it is contributed.
Collaboration between multiple advisers
In the current healthcare system, each doctor and lab and adviser makes notes in their own records. They don’t communicate with each other, and patients don’t get a complete discussion and recommendation. We can solve this problem by inviting advisers to the same HumanDB.
A HumanDB will be free and open source. We have a big job ahead of us, and we need to get the maximum number of people working on it.
We will use a separate database for each person. The current healthcare system uses separate databases for each organization, and one database contains fragments of information about thousands or millions of people. These databases are useful for research and billing. However, they have many disadvantages for humans: software quality problems, security leaks, incomplete records, limited data size, limited accessibility, limited apps and analytics, barriers created by data privacy concerns, barriers between nations, and barriers between advisers.
We will gather data about a human in one place. Our goal is to make data useful for advisers, apps and analytics. It’s most useful when they have access to all of the data in one place. The US healthcare industry has invested billions in a different approach, called “interoperability”. This theoretically will enable any healthcare provider to request data from any other healthcare provider when needed. The crypto bubble produced a lot of schemes to do the same thing, while adding a blockchain to track the data locations and access permissions. These schemes don’t work, because many problems arise when they go to get the data. And, they can’t handle the big data that we want to use.
Code containers make it possible
A HumanDB is built from components that are packaged into “Docker containers”. We can deliver containers carrying new apps and upgraded software, wherever your HumanDB is installed. Related software called Kubernetes can deploy a fleet of millions of HumanDBs into a cloud datacenter.
Container management software like Docker and Kubernetes is important for big companies. In the past, these companies needed years to upgrade a software application. With container technology, they can mix and match containers to get the latest software capabilities, and they can upgrade a container in minutes. Amazon and Google have used this software infrastructure to dominate the modern economy.
HumanDB deploys similar software for one human.
Where we are now, in May of 2019
Over the last five months, Jordan Denison and Andrew Popp have been building a prototype of HumanDB. The container architecture works, and developers can install it and build upon it. We are about to release it into open source. It doesn’t have apps and analytics yet. It does gather and host data. It collects various types of genome sequences in files. It collects traditional medical records a personal FHIR server. It provides a useful gathering point for experts and data, and a foundation for developing apps.
The interaction with experts is low tech, but important. You can invite experts to a discussion forum. Patients will often consult doctors and experts that are in different places, and these experts typically don’t talk to each other. We encourage collaboration by posting reports and recommendations to a discussion thread. Then we can invite all of our experts to comment on the same thread.
The terrible truth about personal health records
Personal health records like HumanDB have a long history of rejection and failure. We’ve known for decades that we will need some type of personal health record in order to provide a lifetime of health advice. In that time, hundreds of PHR projects have started and failed. I interviewed people from nine of these failed projects. They ranged from big company efforts like Google Health and Microsoft Healthvault, to startup operations like Dossia. I learned that there is no demand for personal health records because persons don’t want them. Humans don’t understand data, they don’t see value in it, and they don’t want responsibility for it. Doctors also exhibit little interest in data. Doctors gather information by looking at the patient in front of them.
Apps and analytics have changed the market. They understand data, want data, and use it effectively. They convert data into advice that humans want. Advice is the desirable product.
Access to data has also improved. Every week I talk to people that say “you’ll never get the data.” These people have spent years in a frustrating struggle to pry patient records out of hospital systems. We haven’t experienced that frustration, for several reasons. We don’t need hospital records because we get a lot of data directly from sequencing and diagnostic labs. We can also get information directly from apps and devices. Fortunately, we are getting data from hospital systems, because many of them have turned on FHIR APIs in the past year. We represent the actual human owners, rather than vendors like Google and Microsoft, who can legally claim the data with no privacy restrictions.
Apple Health is one PHR project that is actually working. It’s working for two reasons. First, it supports apps that add value. Second, Apple does not take personal data into their own databases. Instead, they provide each customer with a separate private key and encrypted data store — a topology that is very similar to HumanDB. This has given them an opening to push for API access, in the name of their customers rather than Apple. Unfortunately, they limit themselves to serving only Apple device buyers (about 12% of the market, which is a good start, but a shrinking share). Apple tightly controls and limits the cloud environment, which is where we can add a lot of processing, globalization, and knowledge. Projects like HumanDB have an opportunity to use the data access that Apple has been lining up, and make it available to more users and more contributors.
Personal control, selling data, and blockchain are mostly noise
A crop of new companies say they will help people control their personal data, or make money by selling it. Taking jargon from this article, “In the future with scalable blockchain architectures, competing self sovereign identity systems will help people take back control of their data as a potential stream of income.”
“Taking control” is exactly the task that people have rejected in over 100 failed PHR projects. They don’t want the responsibility and they don’t see the value.
That this will generate a “stream of income” turns out to be a fantasy. The market value of the data from a $1000 sequencing run is about $20. This price is declining as the supply of donors increases.
The idea that we need to sell data is deeply flawed. It’s based on the idea that vendors should know more about us so that they can sell to us and manipulate us. It’s a push model where they figure out how to sell us stuff they want to sell, rather than a pull model where we figure out how to buy what we want to buy. A HumanDB user can employ private data and personal digital advisers to shop and buy more intelligently. Our digital agents can represent us on the Internet, finding and procuring products and services that meet our goals and extend our life. This does not require handing over data, or selling data, or sitting through sales pitches.
If people can use data to directly get valuable advice and services, they won’t care about selling data to get $20 from a drug company. They will care about contributing to research. We are collecting a lot of data that we don’t know how to interpret. Researchers learn how to advise patients by pulling together data about a lot of human cases, and looking for patterns. We can make it easy for HumanDB users to contribute data to research databases.
“Blockchain” is popular with personal health and data projects because they think they can make money by selling blockchain tokens. Unfortunately, blockchains are terrible for keeping track of big, personal data. Blockchains are publicly shared databases. The shared data gets replicated thousands of times to participating “nodes”. So the data must be small. What gets stored in a blockchain is actually pointers to encrypted data that is saved somewhere else, such as a HumanDB.
Blockchains can organize a market around open source software like HumanDB. But, nobody knows how this will work, and all of the proposals so far have been ineffective. Until we figure it out, we can support this project with a much simpler business model — an appstore.
Apps, experts, and data
We will improve the flow of medical knowledge by creating an open global marketplace for apps, experts, and data. We can imagine it as a store. HumanDB is free, but the people who provide apps, experts, and data into a HumanDB will often expect to get paid.
The current market for medical knowledge is very fragmented, and often expensive. Each regional cancer center has a specific set of tests and treatments that they understand and can deliver with quality. This lags behind the rapidly expanding set of scientifically valid tests and treatments. When they say a patient has “terminal” cancer, they already know their process won’t work. Shopping outside the cancer centers is confusing. Each vendor of tests, drugs, trials, etc. has a specific thing that they want to sell. These vendors will say “buy X”, and they won’t help you sort through X, Y, and Z.
We will need a buying guide to the various types of data and diagnostics.
There are a lot of different tests. Consider the subset of tests that use genome sequencing. “Genome sequencing” actually describes wide range of diagnostics. There are panels (sets of up to 500 genes that are designed to make specific treatment decisions), whole genome and whole exome sequencing at various levels of “coverage” and “depth”, and RNA sequencing. These can be applied to a human, or a cancer, or both. Similar tools can do “shotgun” sequencing to figure out what bacteria live in your gut. New technology is bringing us “liquid biopsy” (an analysis of trace DNA in your blood) and “single cell sequencing”, since cancers often have multiple cell types.
Even professionals need a guide to:
- The conditions and triggers that call for more diagnostics
- The available diagnostics, and the healthcare recommendations that might come from each diagnostic
- The cost and insurance coverage
- How to order the diagnostic
- Connecting to follow-on advice. This includes capturing information, including the raw data, and the resulting report with recommendations, in a HumanDB or similar record. Then we can can share it with more experts and new analytical apps.
We will want a catalog of experts that can interpret this information. Experts may be doctors and advocates that work with patients. Or, they may be experts in specific kinds of analysis and treatment.
Then we can add apps. I see a rich ecosystem of apps that are ready to become better informed by using something like HumanDB.
Here are some of the emerging categories.
Expert systems that recommend drugs, trials, and treatments are a growing category that will improve in scale and quality. These products typically produce a report with recommendations in a sorted list. The list usually includes drugs, and sometimes includes clinical trials and other treatment recommendations.
The largest category of reports comes from labs. A sequencing lab or medical testing lab will usually deliver a report after they get data. This report will contain a list of generally accepted biomarkers, “oncogenes”, and treatment recommendations. The good reports link to research papers that explain the findings. Some labs do this manually, with spreadsheets. Some labs, such as Foundation and Tempus, have done a lot of research to develop their own software and analysis.
Regional cancer centers and tech companies are also making these products. This category is fragmented. It doesn’t make sense that each regional cancer center and each lab is doing their own analysis. A patient wants a comprehensive report, from a world-class provider. We will see unbundling, where the lab work and data extraction is separated from the analytics. And, we will see analytics consolidated into a small number of providers that can offer comprehensive recommendations, and continuously improve them.
IBM’s predicament shows how a better data architecture will produce better results. IBM has three different products which each produce a fraction of the advice a customer would want: Watson for Genomics, which looks at genome sequences, Watson for Oncology, which looks at hospital data, and Watson for Clinical Trials, which matches hospital data to a clinical trials database. If they could get a unified view of their human customers, they could provide more comprehensive advice. All of the expert systems that I looked at can be enhanced by using a broader range of data.
These expert systems are built of rules that say “if you see this, recommend that”. We will eventually have a community process to build and test a comprehensive rule base. For inspiration, we can look to the process of recommending clinical trials, which the US government has organized by asking trials to register at ClinicalTrials.gov. It currently lists 50,266 studies that are recruiting patients. This information powers trial recommendations from many apps and advisers. In a future article, I will cover the design of a community repository for treatment rules.
Analytics for personally engineering treatments will become increasingly important and diverse. I presented two examples in my logo wall. Advaita provides and application that helps a human expert drill down on cancer genetic variants and metabolism. In principle, this will help them to discover vulnerabilities that could be attacked by drugs that were designed to affect a particular metabolic process. Neon Therapeutics makes neoantigen vaccines, and they have built software that will qualify patients and design the vaccines.
DNA analysis for consumer wellness is available in the form of apps through from sites like Sequencing.com and Helix. They draw apps from dozens of vendors, and the category is growing fast. They offer apps in categories like “health”, “ancestry”, “fitness”, “nutrition” and “beauty”. Sequencing.com shows that interoperability can work for genetic analysis, by accepting data from many sequencing labs in many formats, and offering a “universal” conversion to the format their apps need. These apps typically download data to their own servers, rather than doing analysis locally.
The current generation of apps provide only a small number of impactful healthcare recommendations, partly because the FDA discourages them from giving medical advice, and partly because the state of the art is not very advanced, and partly because they are targeting a market of currently healthy people. For medical use, we will need to increase the “depth” and accuracy of the sequencing, bring in experts to deliver real medical advice, and expand the range of contributors to the state of the art.
Wellness packages are new services being pioneered by several startups I have talked to. They will measure you with wearable devices, or ask for monthly blood samples. Then they watch for problems, and tell you how to eat right and be healthy. Their goal is to wrap this up in a nice little subscription bundle. Our goal is to expand the bundle.
SMART on FHIR apps are an emerging class of Web and mobile applications that connect to standard health data sources. They can connect directly to the HumanDB FHIR server.
Mobile apps by the thousands are helping consumers navigate their health, wellness, and healthcare. These apps often connect to a FHIR database to learn more about their customer. They don’t often share data, which is a new opportunity they will get from the personal database.
Driving down costs
Tests are expensive, and new cancer treatments are very expensive. We have an obligation to drive down costs and increase access. Technology improvements and volume increases will result in price reductions. Globalization can help us drive the process. Here are some tactics:
Unbundling. We can separate the diagnostic lab from the analytics providers. This helps us increase volume and drive down costs on both sides. We can separate more of the components of treatment into a chain of best-of-breed suppliers.
Transparency: Costs will be lower when consumers can see costs and shop effectively.
Expanding the target cohort. “Precision” treatments make medicine more expensive by seeking small cohorts that will respond to a particular drug. A “Personally engineered” approach, such as a neoantigen vaccine designed for a single person, may work for a wider range of cancers.
Concentrating volume. When we create a global marketplace, we create the opportunity to concentrate volume in one place. This can drive down costs.
Control your future
In the future, software advisers will know you at a molecular level. You will want these advisers, because they will give you new superpowers, and they will extend your life. It is important to ask whether these advisers will be under the control of your hospital, your government, a big Internet company, or whether they will run on a HumanDB that is under your direct control.
This is the second article about HumanDB, a global marketplace of apps, experts and data for longer life. These apps, experts, and data will understand our human users at a molecular level, find the best available treatment options, and equip an expanding universe of care providers.