The use of artificial intelligence (AI) is enabling researchers to undertake advanced molecular testing, which can be developed in record time through assays built using the latest in synthetic biology, AI and CRISPR technologies.
CRISPR technologies are gene-editing tools that enable precise modification of DNA, offering unprecedented capabilities for genetic research.
The integration of AI and CRISPR technologies has the potential to contribute to the rapid creation of new diagnostics, particularly in the context of pandemic preparedness.
Sherlock Biosciences, the first company to market with a CRISPR-based test for COVID, has developed a diagnostic tool that can detect the unique genetic fingerprints encoded in virtually any DNA or RNA sequence in any organism or pathogen.
The platform is highly flexible and can be applied to any organism which has been sequenced, including emerging diseases. Furthermore, it has the capability to identify slowly mutating sequences, even in rapidly evolving pathogens such as viruses.
This provides a future-proofing aspect to diagnostic designs so they can continue to be accurate years down the road. The primary use case of the system today is to enable the rapid design of the nucleic acid components which power diagnostic tests.
“We accomplish this design process through two major computational abilities: our bioinformatics and machine learning pipelines,” says Leland Hyman, a machine learning scientist with Sherlock.
The bioinformatics pipeline searches vast genomic databases to identify target sequences which will provide the greatest sensitivity and specificity.
“We pair that with our machine learning pipeline, applying deep learning models which predict assay performance based on the physical characteristics of the primers and the sequences being detected,” Hyman says.
This allows the company to go beyond human capabilities and make sense of the complicated, microscopic events occurring in a test tube. The result is a set of optimized assays which are then tested in the lab.
“By streamlining the development of diagnostic testing, AI enables us to respond more rapidly to emerging pathogens,” Hyman says.
The specific advantages and capabilities these technologies bring to the development process include reducing the amount of experimental assay development work needed.
“We can focus on a smaller set of primer candidates — specifically those for which the system predicted a higher probability of success,” Hyman says. “Additionally, the machine learning models can learn from the primers we do test experimentally, so the accuracy of the model improves over time.”
The computational tools are intended to be used by scientists in the lab with little to no programming experience.
“They do not need to understand the inner workings of the bioinformatics or ML pipelines,” he says. “They can instead rely on the system just as another tool in their arsenal.”
He explains single DNA and RNA mutations can be difficult to detect using common amplification methods such as PCR – the difference in sequence is simply too subtle to pick up.
“Fortunately, CRISPR systems have evolved to have excellent sensitivity to even small mutations,” he says. “By combining DNA or RNA amplification with our CRISPR technologies, we can therefore detect mutations with exceptional accuracy.”
Previously, scientists developing a new diagnostic assay had to test hundreds of sequences in the lab, leading to longer development times and consuming a large amount of labor and reagents.
“This technology allows us to narrow that down to a much smaller subset to enable rapid and streamlined diagnostic development,” Hyman says. “The reduced R&D cost ultimately helps to enable a more affordable and globally accessible diagnostic.”
He explains bioinformatics processes and access to genomic databases are both indispensable when it comes to diagnostic assay development.
“Genomic databases allow us to find virtually any example where a given pathogen has been sequenced from a clinical or environmental sample,” he says.
By examining the mutations found in all these samples, the bioinformatics pipeline can identify consistent and slowly mutating target sequences even in rapidly evolving viral genomes.
“This helps us to avoid any nasty surprises and delays during assay development,” Hyman says.
For example, there could be a case where a flu primer set might work well in initial tests, but not on real flu samples later in the assay development process.
“Without our bioinformatics pipeline, these types of cases might force additional rounds of R&D to be performed and greatly inflate both timelines and budgets,” Hyman notes.
