A group of researchers from the University of California-Irvine and -Riverside have recently presented a technology that can successfully read sound patterns of a DNA synthesizing machine, raising questions of its pros and cons.
Researchers from the University of California, Irvine and University of California, Riverside recently presented a paper at the Network and Distributed System Security Symposium highlighting the potential threat of the sounds emitted by lab instruments and machines. To address rising biosecurity concerns, the group devised a machine-learning algorithm that can successfully reconstruct what a lab instrument was used for through sound recordings.
"Any active machine emits a trace of some form: physical residue, electromagnetic radiation, acoustic noise, etc. The amount of information in these traces is immense, and we have only hit the tip of the iceberg in terms of what we can learn and reverse engineer about the machine that generated them," said Philip Brisk, a UC Riverside associate professor of computer science who worked on the project.
To prove this, the group tested the technology on a DNA synthesizer machine.
A DNA synthesizer machine can build custom DNA molecules using basic ingredients. This is typically used by researchers to construct segments of DNA that will be inserted into an organism’s genome. They are also often used to create pharmaceuticals and other similar products. The tubes and chambers through which chemicals flow create the distinctive sound of the machine.
They were curious as to whether it was possible to determine what it was producing through the sound of its components during the manufacturing process.
All DNA is composed of four bases: adenine (A), guanine (G), cytosine (C), and thymine (T), which can be arranged in infinite sequences. The researchers speculated whether they could identify the sequence of bases in the DNA synthesis.
The members set up microphones, which were similar to the ones in smartphones, in spots near the DNA synthesizer in John C. Chaput’s, a professor of pharmaceutical sciences at UC Irvine, lab. Even by filtering out background noise and adjusting the recording, it would still be too subtle for a normal human to make out the difference of each sound.
"But through a careful feature engineering and bespoke machine-learning algorithm written in our lab, we were able to pinpoint those differences," doctoral student, Sina Faezi said. The algorithm helped the researchers easily distinguish when the machine produced A, G, C, or T. The software was able to identify the correct DNA patterns with 86 percent accuracy. The researchers ran it through further DNA sequencing software, which successfully boosted the accuracy to 100 percent.
The researchers stressed how this method could raise potential biosecurity concerns. While this method could help identify serious pathogens, such as anthrax and Ebola, and prevent bioterrorism, it could also allow terrorists to intercept biological secrets.
"A few years ago, we published a study on a similar method for stealing plans of objects being fabricated in 3D printers, but this DNA synthesizer attack is potentially much more serious," said Mohammad Abdullah Al Faruque, a UC Irvine electrical and computer engineering professor.
After witnessing the success of their technology, the researchers recommended that labs using DNA synthesizing machines integrate security measures. Suggested measures include enforcing strict control access to machines and removing recording devices near them. Additionally, they also encourage manufacturing companies to design machine components that make less sound— through a complete redesign, repositioning of parts, or the use of sound absorbers.
According to both Brisk and fellow UC professor William Grover, the method can be easily applied because every machine emits a distinguishable sound. "The take-home message for bioengineers is that we have to worry about these security issues when we're designing instruments," Grover said.
The research was supported in part by National Science Foundation awards to Brisk and Al Faruque.
For more stories like this, check out how Computers Use Machine Learning to Detect Radiation Damage Better Than Humans Do.