NEW YORK – While PCR may be an old dog, that doesn't mean it can't learn new tricks. A research team in Australia recently proposed an artificial intelligence-based method to improve PCR speed or quality by altering the conditions as the reaction progresses.
The researchers are currently applying a version of the method to forensic testing, but they are also seeking commercial partners to potentially develop an instrument and other applications.
Adrian Linacre, the head of forensic DNA technology at Flinders University in Adelaide, is corresponding author on two studies published last month in Genes about the method, dubbed Smart PCR.
While the use of PCR to analyze short tandem repeats (STR), mtDNA, and SNPs has evolved rapidly, its evolution in the forensics space has slowed even despite continued advancement in clinical fields. One point where forensics advancement has stalled is the processing of tiny trace, or touch DNA samples, according to a recent review coauthored by Linacre.
In forensics, trace DNA samples from materials that a person has touched are the most common type of samples submitted to the lab for testing. Unfortunately, the success in amplifying STR data from such samples is very poor, Linacre said, such that many samples collected at crime scenes fail to generate any DNA data.
This can be related to low amounts of DNA deposited at the scene — trace, or touch DNA, comes from as few as eight skin cells — or due to the presence of PCR inhibitors like soil or blood that limit amplification. But, much as is the case with tumor biopsies, the tiny starting samples are precious, and a lot hangs in the balance of a successful PCR reaction.
STR analysis typically involves endpoint PCR of a few dozen signature sites, and commercial kits like Thermo Fisher Scientific's GlobalFiler and Qiagen's Quantiplex Pro are widely used.
However, Linacre noted, "currently the PCR process in forensic science is invariant" even despite minor changes that can be made to the recommended cycling conditions of commercial kits.
As is typical, the PCR program is also invariant within a run, he said, such that the same cycling parameters are used at the start, when there is little DNA and the enzyme is most active, and at the 28th cycle, "when there are billions of molecules and the enzyme is becoming inactive," Linacre said.
In theory, an adaptive program could reduce the extension time in the initial cycles and increase it as the template increases and enzyme processivity decreases. Or, denaturation temperature could be decreased if amplicons are short.
And so, Linacre and his colleagues set out to build a system that could be easily varied using machine learning and feedback loops.
"Our concept is to monitor the PCR process and adapt the time and temperature at the denaturation, annealing, and extension steps," Linacre said, by monitoring the doubling of fluorescence at each cycle and adapting parameters based on this feedback.
As described in the proof-of-concept study, the team is using Chai Biotechnologies' open-source PCR machine, dubbed Open qPCR, to test the process. This potentially allowed the PCR program to be adjusted "cycle by cycle," Linacre said.
The researchers also presented their approach for defining PCR goals, scoring the performance of the system toward achieving those goals. For example, one goal could be to improve the quality of the DNA profiles, including measures like having fewer artifacts, more complete DNA profiles, or better balance between peaks. Improved PCR effectiveness could also be gauged by decreasing the overall PCR program time without impacting the quality of the DNA profile. Yet another goal might be increasing efficiency by lowering the sample volume required or the cost of reagents needed.
With these parameters defined, the researchers connected a dual-channel Open qPCR instrument to a web interface, giving them the ability to interact with the thermal cycler's application programming interface (API) using a custom JavaScript program and further demonstrate the ability to dynamically control PCR conditions.
However, although the team described the theory of how they could make on-the-fly changes to PCR conditions, Chai did not have the ability make changes while a program was running.
In a second study, the team demonstrated a large-scale PCR cycling condition alteration experiment using an AI algorithm called firefly. Here they found that AI-generated modifications saved 30 minutes while producing DNA profiles with higher scores than those from the standard GlobalFiler PCR protocol. The team also showed firefly-based Smart PCR yielded similar results in a poster presented last month at the meeting of the Congress of the International Society for Forensic Genetics.
Digital PCR is another technology that could potentially address forensic lab struggles, as it typically tolerates inhibitors better than qPCR and can excel with rare targets. Qiagen recently signed an agreement with the US Federal Bureau of Investigation to develop and evaluate a QiAcuity assay to simultaneously quantify nuclear, mitochondrial, and male DNA, for example. Companies like Othram and Verogen, an Illumina spinoff acquired by Qiagen last year, are also applying sequencing technologies to forensics. And, Bode Technology research recently teased the possibility of using direct PCR from small samples.
Going forward, while the Flinders team is currently developing its methods on the commercial qPCR machine, ideally, it would like to build its own dynamic instrument.
Adapting the Smart PCR process from forensics to clinical or other applications will be straightforward, however, since "the same principles apply to any PCR amplification process," Linacre said. The team also hopes to commercialize the Smart PCR technology eventually. "We can take it only so far, and then wish to partner with a company who could take this to production," he said.