CovSort

Inspiration

The search for carriers of COVID-19 is done primarily through testing using RT-PCR's. These tests are the most common way to empirically identify carriers of the virus, and are today conducted "one by one" – every sample from every patient is tested separately. This current testing method is problematic due to the following reasons: 1) The number of patients and samples gathered today supersedes the number of tests that can be conducted daily.

2) A shortage in equipment and resources denies the ability to increases tests, moreover when such equipment is in world demand.

3) Even for targeted populations, most patients receive negative results.

As a result, the testing system today is at full capacity, but falls short of the need. A new review, published by Prof. Roy Kishony from the Technion,Israel, has shown that it is possible to evaluate tests for COVID-19 taken from multiple patients. By combining samples into one "pool", one can achieve an accurate result for that pool, As shown in the following article: https://kishony.technion.ac.il/wpcontent/uploads/2020/03/2020.03.26.20039438v1.full_.pdf

This new tool is powerful, but we have shown that for the current rate of positive samples (~8%) identified in the population, creating random pools of samples will not be sufficiently efficient - our project solves this problem to a large extent

What it does

By using data received from the last month of testing, we created an algorithm based on machine learning, that predicts the probability of a patient being negative or positive for the virus. This prediction will enable us not only to identify high risk patients, but also to "pool" low risk samples to one test, allowing swift ruling out of all patients if the mutual test is negative. Our model can accurately predict 95% of the negative patients from recent tests, and 85% of them can be predicted with less than 2% chance of error.

How we built it

first, we drafted the general probability equation for finding a positive sample. after that we introduced pooling sizes to calculate the optimum pool size for reducing test numbers, in regard to the probability size. using this knowledge, we tried several approaches to separate samples: from binary tree separation to naïve codes of separation. we saw that a neural network works best regarding results, but also is the easiest system to implement now to testing lab protocols.

Accomplishments that we're proud of

we have received praises from faculty staff at the Technion, who are now guiding us with this new system. we have been approached by several hospitals and investors for implementation of our idea. moreover, our idea has been pitched to the Israeli health ministry and to the Israeli prime minister's office. we believe that the results our system provides are of high standard and low error, which we see as a great personal accomplishment

Challenges we ran into

it was very hard to acquire data regarding tests, results and information of the patients. many data sources are lacking and the motivation behind them is mostly unknown. although we acquired a lot of interest, actual implementation of our idea has not occurred yet, and needs more push above what we can provide. we have also identified a discomfort of physicians with using pooling tests, with good reason. we believe that pooling is one future improvement which most be, at least, evaluated.

What we learned

we learned a lot about how the testing of covid-19 works, from the science of PCR machines to the testing protocols of first aid recruits. we learned how to think not like engineers, but like lab technician, physicians, patients and even like business man, all as part of our desire to improve our reasoning and presentation of our idea

What's next for CovSort

we whish to implement our system in labs and conduct a "proof of concept" test to show effectiveness and accuracy. after that, if there is interest, to continue implementation in more labs with government support. this software is not only of value for the corona crisis, but for any outbreak where the majority of testing is conducted via PCR machines