Interview written and condensed by Tara Luther, Promega

Despite significant technological advancements over the past few decades, the success rate of generating DNA profiles from sub-optimal samples remains low. While considerable progress has been made in the DNA recovery, extraction and profiling processes within forensic science, there has been relatively little focus on optimizing the fundamental PCR process that underpins these steps. Recognizing this gap, we set out to enhance PCR efficiency (and profiling success) by re-evaluating the process at its core. Our research is specifically aimed at optimizing PCR conditions to improve DNA profile quality and success rates, particularly for challenging trace, inhibited and degraded samples.

Traditional PCR methods used in forensic DNA profiling follow a standardized, uniform cycling approach where reaction conditions remain fixed throughout the run. However, the chemical environment within a PCR tube is dynamic–reagents are depleted, and enzyme activity declines as amplification progresses which significantly affects efficiency. This is particularly problematic for sub-optimal samples, which are often highly complex and contain low template quantities, degradation and inhibitory compounds. The standard protocols for DNA profiling do not account for these variations within the tube, which may be contributing to poor quality DNA profiles from such samples.

Our smart PCR system aims to overcome these limitations by dynamically adjusting cycling conditions in response to real-time feedback from the reaction itself. By tailoring the process to each sample’s unique characteristics, this approach has the potential to improve amplification success rates and enhance DNA profile quality, particularly for challenging samples that currently yield limited or non-informative profiles.

The ability to monitor PCR progress in real time and adjust cycling conditions in response is the goal we are working towards. In that pursuit we have developed a comprehensive databank of DNA profiles that characterizes the impact of altering specific elements in the PCR process, such as changing the denaturation timing, on the quality of the resulting profiles. It also highlighted the key features of the profiles affected by each change, such as allele balance and peak heights. This step was critical in the development of the model because, like any machine learning system, the quality of the dataset directly influences the accuracy and performance of the algorithm.

Once the databank was established, we developed a machine learning algorithm that used the PCR cycling conditions (parameters) as inputs and the features of the resulting DNA profiles as outputs. Using the training dataset, the model was taught to associate different cycling conditions with the quality of the profiles produced. Any PCR programs that produced “good” quality DNA profiles were attractive to the system, and programs that produced “poor” quality profiles were repulsive. Using this approach, our system is able to suggest cycling conditions that will improve amplification and ultimately DNA profile quality.

Conventional static PCR processes, which are commonly used in forensic practice, require analysts to wait until the end of the amplification run and rely on downstream processing (such as separation using Capillary Electrophoresis and data analysis in GeneMapper) to assess whether amplification has been successful. In contrast, our system uses real-time fluorescence feedback to monitor the amplification process as it unfolds and aims to provide immediate insights into the reaction's progress. This real-time monitoring focuses on amplification efficiency, which serves as a proxy for assessing the effects of changing reaction conditions within the PCR tube.

Through our research we've identified specific alterations to cycling conditions that can enhance amplification efficiency. By integrating this knowledge into the system, we can dynamically adjust the PCR conditions during the run based on the real-time feedback. The system continuously monitors the amplification efficiency and can adjust the cycling conditions to optimize the process, which is something that the current static PCR methods cannot do. This capability would allow analysts to see whether the PCR is proceeding well in real time, offering an immediate opportunity to intervene in the process to optimize the genetic information that can be obtained.

In isolation, the time saving from an AI-driven PCR system may not seem particularly significant. However, when considering its broader impact on forensic workflows, the potential benefits become much more substantial.

A smart PCR system has the capability to consolidate both qPCR and endpoint PCR for DNA profiling into a single process, streamlining operational workflows. Additionally, by optimizing cycling conditions, it can accelerate amplification for ideal samples while also improving success rates from suboptimal ones. This, in turn, reduces the number of samples that would require repeated processing or ultimately yield no useable results which would save both time and resources.

While this technology is not intended as a direct replacement for rapid, field-based DNA analysis, it offers a powerful alternative for laboratory-based forensic casework. By increasing efficiency and improving overall profile quality, a smart PCR system has the potential to enhance sample throughput and reduce bottlenecks in the processing of challenging samples. There are many applications for a smart system outside forensic science and given the sheer volume of work that relies on PCR, even a very modest improvement could have very significant overall impacts.

The use of AI in forensic science raises interesting challenges, particularly in the public and legal arenas. AI is often perceived as a black box making unchecked decisions, but when implemented transparently and responsibly, it is simply another means of improving forensic processes. Therefore, gaining public and regulatory acceptance is key. Transparency in the methodologies used along with ensuring reproducibility will help build confidence and acceptance in forensics, just as with other technological advancements in the field.

A crucial step toward operational implementation would be sufficient trials to meet general acceptance by the forensic science community and move towards validation to allow accreditation. The Australian Government has also proposed AI guardrails to guide implementation of AI in high-risk settings, including forensic science, by providing a framework for meeting regulatory and ethical expectations. Aligning any AI systems with these guardrails will likely help with integration into accredited workflows.

Additionally, exploring compatibility with existing validated and commercially available forensic kits may help streamline adoption. Aligning this approach with established workflows could help expedite acceptance and implementation, making the transition to a smart PCR system more seamless within forensic laboratories.

We have explored the compatibility of smart PCR with existing validated and commercially available systems in forensic laboratories. Our findings indicate that integrating a smart PCR process into these existing systems could indeed be done. The real challenge, however, does not lie in adapting existing systems, but rather in ensuring that reliable, reproducible, and validated methods can be implemented consistently across different laboratories. This is a common hurdle faced by any new method introduced to forensic practice and is not unique to our work.

That said, our research has demonstrated that a smart PCR system is not strictly necessary for improving PCR processes. Early indications show that there are adjustments that can be made to existing PCR workflows that can potentially enhance profile quality or reduce runtimes for certain samples, independent of the AI aspect. For example, altering cycling conditions across a PCR run to better account for the changes within the reaction tube can be implemented on current PCR machines without the need for AI. This was actually how we initially tested the concept.

So, to circle back to the question, yes, I believe this process could be integrated into existing systems in a number of ways and in doing so can ultimately enhance both the quality and efficiency of forensic DNA profiling.

AI-enhanced PCR has potential applications far beyond forensic STR profiling. Within forensic science, it could be applied to areas such as microbial forensics, environmental DNA analysis, and RNA-based research. However, beyond forensics this technology could transform many other PCR-dependent fields, including medical diagnostics, biotechnology, and environmental monitoring. Of these, I believe medical diagnostics is the most likely to adopt AI-enhanced PCR first.

The medical field has a pressing need for high-throughput, accurate, and efficient PCR-based testing, particularly in infectious disease detection, oncology, and genetic screening. Any optimization of this process could significantly reduce reaction times while maintaining or improving sensitivity, which is critical in clinical settings where rapid and reliable results directly impact patient care. Additionally, biotechnology companies are continuously developing new PCR applications, and integrating AI could help improve reaction efficiency, lower costs, and enhance automation in research and commercial settings.

While environmental monitoring is another promising area, it lacks the immediate demand to drive early adoption. Similarly, forensic implementation may take longer due to the rigorous validation and legal requirements for casework use. Given these factors, I feel that medical diagnostics is the most likely field to lead the adoption of AI-enhanced PCR, with biotechnology following close behind.

Our current focus is on collaborating with forensic providers to transition our smart PCR system into real-world casework settings. This includes assessing how the system performs with the diverse range of samples encountered in forensic laboratories. Additionally, we are investigating opportunities to refine the system further, ensuring that it meets the stringent requirements of forensic casework. Beyond STR profiling, we are also actively exploring potential applications within forensic science and in other PCR-dependent areas where the ability to dynamically optimize PCR conditions could have a profound impact.

Looking more broadly at AI-driven technology, one major improvement I’d like to see is greater accessibility to open-source AI tools. Currently, developing unique AI-driven systems faces numerous obstacles, from proprietary software limitations to restricted datasets, which make the development process more challenging. A shift toward more widespread open-source, customizable frameworks would accelerate innovation and make it easier for researchers to use AI in their own work. Such a movement could help drive the AI revolution forward, pushing both forensic and broader scientific research into the future.