Researchers at the University of Toronto have made an advancement in DNA-based neural networks, paving the way for more portable and scalable molecular computers. By implementing new strategies to overcome existing limitations, the team has developed neural networks that can process information quickly and accurately without relying on traditional electronic components.
This research was published in the latest Journal of American Chemical Society issue: https://pubs.acs.org/doi/10.1021/jacs.4c07221.
Neural networks are powerful algorithms for processing complex datasets. While they are typically operated using computers made from silicon, DNA-based neural networks offer an alternative by enabling operations directly on biomolecular inputs, such as DNA and proteins. Additionally, these networks can function in environments where electronic devices are impractical. However, the challenge has been scaling up such DNA computers. Traditionally, creating a single neuron in a DNA-based network required numerous nucleic acids, leading to long assembly times and increased errors in information propagation. The researchers aimed to solve this by simplifying the architecture, making it possible to build more neurons with fewer components.
“DNA computers can be powerful for applications such as point-of-care diagnosis in low-infrastructure settings and molecular digital data storage. However, their computing power has lagged due to limited scalability” said Dr. Leo Chou, an assistant professor at the University of Toronto and the corresponding author of this research.
The research team employed several strategies to improve the DNA-based neural networks: they used enzymatic synthesis to produce high-purity neurons and reduced neuron crosstalk by organizing them into spatially segregated clusters. The researchers devised a neuron design that enabled them to encode neuron connectivity using a minimal set of DNA sequences, leading to faster and more accurate neuron activation. With this neuron design, the team was able to rewire the network into various neural circuit motifs, such as cascading, fan-in, and fan-out circuits, showcasing the versatility of their design.
“The neurons we designed can be quickly rewired into completely different circuits, and their assembly and computation steps can be automated with microfluidics,” said Ryan Lee, a PhD student and the lead author. “These features could condense the design and testing phases of DNA circuits from a period of many months to as short as a few days.”
Looking ahead, the researchers plan to explore ways to further enhance the scalability of these networks. “We are interested in modeling more complex computing architectures and experimenting with different chemistries to further advance DNA computing systems. In parallel, we plan to test how these systems can process complex inputs, such as DNA and RNA extracted from blood, for applications such as disease classification” said Professor Chou.
This work was done in collaboration between the University of Toronto and the University of Edinburgh.