wePool Algorithmic Platforms

Shortages in PCR testing currently prevent the containment of COVID-19. With only a small fraction of the population tested, facilities across the world are already overwhelmed. Only symptomatic patients are tested while asymptomatic carriers continue to spread the disease. Consequently, organizations and countries respond based on incomplete data. There is a need to either drastically increase testing, which is currently improbable, or maximize the information obtained per test. 

Pool testing is an innovative solution to maximize information per test, however, disease prevalence is too high to perform pool testing. We propose to use machine learning to predict an individual’s disease prevalence to employ intelligent pooling. 

Intelligent pooling drastically increases testing capacity, allowing us to redefine the rules for testing, and begin to test a broader population. This would have the impact of preventing future outbreaks and allow the world to reopen safely.

On January 1st 2020, there were 27 confirmed cases of COVID-19 in the world and 0 confirmed deaths; as of June 17th 2020, this number has risen to over 8 million cases and 444,000 reported deaths.

The world watched as we collectively struggled to keep up with the rapidly moving novel coronavirus that left no country behind. The US has not been spared, and is in fact the hardest hit country. Several factors led to the rapid increase in disease spread across the country, of which, one of the main bottlenecks was the lackluster testing rate. 

While bureaucratic red tape has been relatively overcome and private labs have ramped up testing across the nation, testing is still insufficient to meet the true demands. The US is currently performing around 400,000 tests daily, which, according to estimates, needs to be increased at least 3 fold in order to perform the frequent and repeated testing required to not only track and mitigate the disease, but also allow safe reopening of the economy. 

This scale of testing is highly improbable with increased test supply alone. Therefore, we propose intelligent pooling, which can drastically increase testing capacity with the same number of test kits.

Our approach to solving testing shortages is to maximize information obtained per test by employing intelligent pooling. 

Pool testing is a strategy where samples from multiple subjects are combined and tested with a single test. A negative test result confirms that all samples are negative; a positive result would require either individual testing or re-pooling samples after breaking the population down into smaller sets. 

Currently, high disease prevalence (i.e. a high rate of positives) prevents effective pool testing: the expected probability of encountering a positive sample within an otherwise negative pool (spoiling this optimization), is high. 

Our approach is to use machine learning to actively minimize this probability by predicting an individual’s disease prevalence, which will be used to segment the population into distinct low prevalence pools. By using information such as age, sex, symptoms, location & contact tracing, we can accurately achieve this, followed by a recommendation on how to strategically pool predictably negative samples. 

Our solution would serve as an intermediary platform between sample collection sites, from which information will be obtained, and the lab testing facility, which receives optimized pooling instructions. 

This way, we drastically increase testing capacity without the need to significantly increase testing supply.

Our vision is to ensure fast and efficient testing availability for everyone. To that end, we want to ensure that all those who need a test are able to receive one, and also provide a tool for frequent mass testing to assure the safety of all individuals returning to the workforce. 

Pool testing is a strategy that could be hugely beneficial in reopening global economies. Additionally, testing facilities have been deeply backlogged and the strain to conduct even more tests is enormous. We have been speaking with hospitals such as MGH and the VA as well as interviewing industry experts in diagnostic testing to understand the feasibility of pool testing given the current testing platform. Pool testing would allow facilities to greatly improve testing capacity with minimal additional resources.

Our solution directly serves organizations thinking about returning to work as well as testing facilities who need an innovative solution to ramp up their testing capability. We are in contact with large and small corporations who are uncertain about how to ensure a safe reopening of facilities. Due to limited testing, they are currently unable to frequently test their workforce to guarantee their safety and prevent an outbreak.

One of the major factors in preventing disease outbreaks is the ability to perform widespread/mass diagnostic testing and contain the disease through frequent retesting of individuals. 

Even after months of tackling the current pandemic, the world is struggling to keep up with testing. 

We provide a computational-based testing strategy that can drastically increase testing capacity with minimal need for additional resources. We believe that our problem and solution fit perfectly with the health security and pandemics challenge as it provides communities around the world a testing strategy based on technological innovation for rapid detection and response to potential outbreaks.

Pool testing is not an unknown approach—indeed it has been used for malaria and even COVID19 in very few countries like Israel and Germany—BUT the application of machine learning technologies to understand how to segment samples in order to pool test efficiently, IS NEW.

Testing demand has continued to rise, outpacing supply. In scenarios where the prevalence rate is already high, if pool testing is used, a need arises to weed out predictably positive samples that may spoil a mix of desirably negative samples.

In other words, by using machine learning techniques, we can predict patient prevalence rates in order to employ strategic pooling: mixing and pool testing predictably negative samples such that the expected probability of at least one positive in an otherwise negative pool (that might 'spoil' the mix) is close to zero, thereby allowing us to reliably clear many subjects in one shot.

It is the combination of an underused method (pool testing) and Artificial Intelligence that makes our approach so innovative, garnering the support and interest from the hackathons' mentors, judges and participants, culminating in our Wins.

Our intelligent pooling solution falls in the category of a new application of an existing technology: this combination of an existing technology (i.e. machine learning) with a relatively novel methodology (pool testing) allows us to inform sample segmentation so that we can pool test more efficiently.

In general, the technology takes data inputs (i.e. de-identified subject data) that is available to sample collection or laboratory analysis facilities. The model then uses Machine Learning to provide a computational testing strategy that outputs an individual’s predicted disease prevalence, and uses it to segment the test population into distinct pools.

The trained model uses weighted features to make this prediction (such as location data, symptoms, etc). Once a prevalence assessment has been established, our algorithm can suggest pooling protocols and pool size strategies.

Our initial simulation results show that our intelligent pooling solution can save up to 70% of test kits used at low prevalence, while at the same time increasing laboratory testing capacity for COVID-19 RT-qPCR by over 300%.

Intelligent pool testing is a novel application of Machine Learning that allows us to understand how to segment samples for more efficient pooling.

Pool testing alone has been used for malaria and recently for COVID-19 testing in few countries like Israel and Germany [Yelin (2020); Eberhardt (2020)]. 

Machine learning techniques are widely used in numerous applications in the healthcare field utilizing AI for optimization purposes [Thomas (2020)]. One recent evidence of a computer-aided prediction model including a trained algorithm concerns ultrasound imaging in breast cancer [Zhang (2020)].

Using synthetic PrecisionFDA data made available to us, we have created a data-ready framework for our model, allowing us to run simulations showing the efficacy of this combination.

Although simple pooling (without AI) has the potential to deliver test kit savings in low prevalence environments, its returns diminish sharply as prevalence increases, making it viable only when the population's prevalence stays within 0-30% [see attached].

At a prevalence rate of 20%, simple pooling could enable 9% savings, whereas with wePool we can deliver savings of 50%, and up to 70% with lower prevalence. Further, our approach enables pooling for populations with a prevalence greater than 30%.

We expect our technology to consistently & actively maintain a strategic prevalence rate of <30%, thereby allowing us to enable a minimum of 40% savings in tests used, and a 300% increase in testing capacity (samples analyzed) at maximum [upper bound] [all simulations were run with a pool size of ≤4].

Shortages in testing capacity have largely impacted the world’s ability to contain COVID-19 and to limit spread of the virus. Facilities across the globe continue to be overwhelmed with only a fraction of the population tested (5% in the US). As a result, only symptomatic patients are tested while asymptomatic carriers continue to spread the disease. Consequently, organizations and countries respond based on incomplete and biased data. 

Pool testing is an innovative solution to maximize information per test. This approach has already proven successful in the current COVID-19 pandemic [Yelin (2020); Eberhardt (2020)], and as of very recently, the FDA is encouraging pooling techniques [FDA (June 2020)]. However, disease prevalence rates still interfere with efficient pool testing. 

Our proposal to overcome this issue is to use machine learning to predict an individual’s disease prevalence in order to employ strategic pooling. Mixing and pool testing predictably negative samples such that the expected probability of at least one positive (that might 'spoil' the mix) is close to zero, allows us to reliably clear several subjects in one shot. 

Our simulations show that this approach has the potential to save test kits and increase test capacity by up to three times current capacity.

This strategy would allow countries and organizations to redefine the rules for testing and begin to test broader populations, potentially saving lives by increasing detection of the virus. 

It is the innovative combination of an underused method (pool testing) and Artificial Intelligence that could have an impact in preventing future outbreaks and also allow the world to reopen safely.

As our technology is currently in the prototype stage, pending training of our model and a clinical trial, we do not yet serve customers.

A year from now, if we were to partner with just two of the lab testing facilities we are currently discussing collaboration terms with, our technology will have been employed for the processing of over 55,000 samples DAILY. This is our basecase assumption, but we expect to work with at least 5 testing facilities.

Five years from now, our technology may become a valuable standard tool for large scale or mass testing, specifically in the context of epidemic and pandemic preparedness scenarios. 

We also envisage our technology expanding from COVID-19 RT-qPCR to additional laboratory tests (antibody testing, next generation sequencing).

WePool’s mission is to provide computational testing strategies for efficient disease management, envisioning ‘Fast and effective testing for everyone’.

Our project recently emerged from the MIT COVID-19 Beat the Pandemic Challenge as a two-time winner from both hackathon rounds. Consequently, we are in an early implementation phase, and are in the process of completing our prototype within the next month.

Our immediate objectives are incorporation and protection of our work product through a provisional patent, so that we can enter in collaboration agreements with key implementation partners. Through this, we hope to train our algorithm with genuine testing data so that we may conduct a clinical trial, and later publish the results in a peer reviewed journal. Within the next year we plan to clear regulatory pathways and launch our first generation technology. Partnering with small and large organizations (e.g. laboratories & hospitals) is crucial to make our technology available for broad use.

The United States will be the first country where we operate, and our operations will expand to countries in Latin America and Europe over the next year. 

We envisage the intelligent pooling technology to be useful for laboratory tests other than COVID-19 PCR (antibody testing, next generation sequencing), and may be an important tool for mass immunity identification as countries begin to reopen.

Five years from now, our technology may become a valuable standard tool for large scale or mass testing, specifically in the context of epidemic and pandemic preparedness scenarios.

Our immediate challenges are of a Financial nature. Having won the MIT COVID-19 hackathon twice, our work has secured initial funding of $5,500. We intend to use this stimulant fund for legal support via incorporation and the protection of our work through provisional patent filing. Moving ahead we will need to secure funds to finalize the wePool prototype, and launch our first generation product.

Technical challenges include access to real world data required to train the algorithm that powers our machine learning technology. We aim to partner with larger organizations such as VA and UCSF for research, algorithm training, clinical trial & publication purposes. 

We envision using this technology first in the US, followed by Latin American and European countries. Healthcare is a highly regulated field, licensing requirements applicable for AI technology are diverse and an area of continuous change. Navigating these regulatory requirements around the world might pose unforeseen challenges and Market Access Barriers. We are aware of current requirements in the US, additional research is ongoing for Europe and Latin America.

Our intelligent pooling technology aims to support efficient disease management. However, cultural differences and healthcare policy requirements may result in data access barriers, interfering with or delaying the implementation of wePool AI. Specifically, access to necessary data to run the algorithm might be problematic in countries where political interests govern lab testing. In addition, some countries will have local language requirements for software, interfering with a fast roll-out.

Financial: Our initial funds include a total amount of $5,500 which we have secured by winning both MIT COVID19 Hackathons. We expect to fund our work initially through sustained donations and grants. MIT Solve is the first of many challenges and funding sources to come, through which we hope to be able to secure funding. In addition, we are actively exploring venture capital & investment options: in fact, there is already interest from a number of non-dilutive Latin American parties.

Technical: We have entered into partnership discussions with University of California (UCSF) and the Boston VA for the following purposes: research, prototyping (through precisionFDA), model training, and clinical trial efforts. We are also speaking with the Broad Institute to understand how the pooling methodology (and the implementation of the wePool technology) would affect the COVID19 testing framework & processes. As a next step, we are working with MIT to be assigned partners (e.g Mass General, John Hopkins, Harvard Medical, among other options) as per our—and the partnering organizations'—needs.

Cultural / Healthcare / Political: We are working hard to build the network required to address cultural and political implementation barriers in our target geographical areas.

Market barriers: We are in contact with experts in the field so that we can be best prepared to achieve the regulatory clearance required to implement our technology as soon as possible.

N/A

Current Full Time Staff: 2 (expected to increase shortly)

Part Time Staff: 1 (expected to increase shortly)

Weekly Collaborators, Volunteers, and Mentors / Advisors: 10 (constantly increasing)

We are an interdisciplinary team of professionals that came together during the highly selective MIT COVID19 Challenge. 

Our team's experience and expertise spans the following fields: Economics, Technology, Business, Pharmaceutical Industry, Biomedical Engineering, Operations Research, Data Science & Analytics, AI & Machine Learning, Bioinformatics, Regulatory Affairs, Clinical Development & Medical Affairs. 

In addition to our core team, we have the support of several mentors & experts in the field who became interested during—and remained engaged after—the MIT COVID19 Challenge. 

In addition to the combined network our team holds—as well as the network we are creating with the collaboration of MIT as consecutive hackathon winners—our team is already well under way in establishing implementation partnerships with the likes of University of California, the Department of Veterans Affairs and the Broad Institute, in order to ensure our solution can be delivered to the full extent of its potential impact.

We are currently in partnership discussions with University of California and the Boston VA with the purpose of gaining access to their clinical test data. With it, we expect to run our model for testing purposes and re-train our technology as needed.

University of California has expressed a desire to conduct clinical trials using our technology, as well as interest in taking part in any publication that emerges out of our research collaboration.

We are also beginning to speak with The Broad Institute, with the purpose of understanding how exactly the pooling methodology (and the implementation of the wePool technology) would affect their COVID-19 testing framework & processes.

Because we have only recently won our second round of the hackathon, we have no other partnerships to share at this time—but we are working with MIT to be assigned partners (such as Mass General, John Hopkins, Harvard medical, among many other options) as per our—and the partnering organizations'—needs.

We are currently evaluating business model options so as to adapt seamlessly into the current testing infrastructure & its changing needs. 

Because the testing process varies between states and private institutions (e.g. collection sites are often distinct to sample analysis sites), we acknowledge that more work needs to be done on this front to understand our potential business model.

Regardless, we can preliminarily propose two models:

One, offering our service as a subscription/Software as a Service model wherein a monthly/yearly fee is charged from the target customer (testing facilities, hospitals, etc.). 

Our preferred model, and one that was recommended by a mentor who runs a testing facility, is a ‘Cost per Dispatch / Pool’ model, wherein wePool would likely be compensated a certain amount, proportionate to the savings incurred by the user, per pool analyzed. 

Such a model would be analogous to already existing software in the Logistics industry, wherein vendors collect fees per warehouse package dispatch enabled by their technology.

We expect to fund our work through sustained donations and grants until we reach a point where our system is ready for widespread implementation. In addition, we are actively exploring venture capital & investment options: and in fact, have already secured interest from a number of non-dilutive sources in Latin America.

Our current roadmap milestones are the following: After we (1) train our model with the aforementioned partner organizations, we expect to (2) complete a clinical trial utilizing our technology. As we achieve expected results, (3) we hope to make a publication widely available prior to (4) widespread implementation of our first generation product.

As the development of our software / technology approaches completion, we expect to incur diminishing costs, and once the product is ready for widespread use, we expect it to produce revenue as per the appropriate business model.

After our consecutive wins at the MIT COVID19 Challenge, the Hackathon’s Co-Director, as well as several mentors, redirected us to MIT Solve and strongly encouraged us to apply.

As detailed previously, we will require funding to cover the operational costs of our initiative. We envision our solution as a potentially important and impactful enabler in the mitigation and containment of the pandemic.

We believe that any monetary grant received through MIT Solve would be an important alleviation of our financial burden, but we also look forward to increasing our connection and affiliation within the MIT environment in order to accelerate our progress and connect with potential partners that can make our solution a reality.

As we are an emerging organization, much of which is aided by volunteers with full-time jobs and obligations, we would appreciate support in the form of expert collaborators, as well as partnerships with organizations that might accelerate our progress.

We are open to collaboration and support on all ends, from the Business Perspective, to Strategic Direction, to Medical Affairs. The goal of our partnership would be to further our roadmap goals: access to clinical testing data, model training, clinical trials & publication review, and lastly widespread implementation.

We are interested in partnering with the Martin Trust Center for Entrepreneurship, as well as related response organizations such as MIT Medical (among others). 

Outside of MIT, we are looking to partner with hospitals and lab testing facilities with the purpose of accessing their data and optimizing their testing strategy (e.g. Mass General, others).

We feel that our technology is a perfect fit as per the prerequisites of the AI for Humanity Prize: wePool AI is entirely powered by—and reliant on—Artificial Intelligence & Machine Learning.

We believe that these methodologies, in combination with our strong Data Science understandings and best practices, are what allow wePool to accurately predict individual subject prevalence in order to enable intelligent pooling, and deliver both substantial cost savings and increases in COVID19 testing capacity. 

We do this to achieve our mission: to provide a computational testing strategy for effective disease management, envisioning fast and effective testing for everyone.

If honored by winning the aforementioned Prize, we would utilize the funds to achieve our immediate goals, and to cover fees related to incorporation and provisional patent filings. We expect any grants received to be able to fund and sustain the development of our wePool AI prototype, as well as cover any costs related to data access, clinical trials using our technology, and filing fees concerning IRBs and regulatory applications.