Business Finland: Finland and Vietnam collaborate to use AI for cost-effective COVID-19 mass testing

Top Data Science (TDS), Vietnam Military Medical University (VMMU) and Ampharco USA started to collaborate earlier this year to utilize AI for high sensitivity, precision and high-throughput solution in detecting SARS-CoV-2 virus and other infectious diseases based on PCR melting data.

The new AI application can find an infected sample from PCR testing when the patient’s viral load is still low. AI, combined with a novel testing method, can scale up testing and radically lower costs.

The number of tests performed to date is substantial. In the EU and the UK alone, over 100 million PCR tests have been conducted this year to detect COVID-19 infections. To prevent the pandemic from continuing to spread, PCR testing requires improvement. We must accelerate testing to diagnose people before they infect others. We need to recognise infections when new viral mutations emerge, and we have to lower the testing costs to handle the epidemic in the long term, especially in more impoverished regions of the world. This is where the TDS AI software steps in.

Researchers at VMMU have developed a method to perform PCR tests to detect the novel coronavirus, as summarised below:

• Novel primers simultaneously target seven regions of the genome of the virus causing COVID-19.
• The cDNA is pre-amplified with a conventional thermocycler and detected by a real-time PCR machine.
• An innovative sample pooling technique allows the pooling of multiple samples into one tube while maintaining sensitivity. When using microtiter plates with 96 wells, nearly 5 000 to 10 000 clinical samples can be tested in each run of a real-time PCR.
• AI-based application software developed by Top Data Science evaluates complex melting data from PCR assays. The results define the samples as positive or negative.

For the analysis, technicians access to the web interface of the application to upload batch data files of the test results. The AI model has been programmed to recognise the melting spectra of positive and negative samples. The virus is detected by comparing the actual test data to the existing data of positive and negative samples. The application currently supports multiple PCR brands, including Rotor-Gene Q, LightCycler, Biorad, Strategene and ABI.

When combined with the VMMU method, the AI analysis of PCR samples provides significant benefits:

• Since several different gene fragments can be targeted by tools to analyse the complex data, the tests are more sensitive. The algorithm can detect a viral load that is 15% of the required viral load in the current standard tests. Infections in patients can be detected earlier than has been possible, even before symptoms start. This helps to contain the transmission of the virus.
• Simultaneous analysis of various fragments can reveal a pathogen even if some part of the genome is mutated. Infections caused by mutations can be detected with this approach when other tests would fail.
• The new testing method with AI analyses is significantly more efficient than current capabilities. Productivity in a laboratory can increase, which allows for mass testing and lower costs. A typical testing centre with one or two real-time PCR machines can perform 3 000–10 000 tests in a working day with the addition of roughly ten conventional PCR thermocyclers (5–10% the cost of real-time PCR machines).
• With improved analysis sensitivity, fewer reagents and materials for viral detection are required, even when testing samples without RNA purification. The same technology allows for the use of more affordable PCR machines in the laboratories as well, significantly improving testing costs.

This article was first published on 18 November by Business Finland.