Graduate Seminar Series: Molecular Stream
Graduate Seminar Series for the Institute of Biomedical Engineering (BME). This day is for molecular stream presenters.
If you would like to invite your Principal Investigator, please add their email via the ‘Add Guest’ button and they will also be notified of your presentation.
Presentation Title: Towards neural-network-on-a-chip: integrating microfluidics with DNA-based computers for powerful point-of-care diagnostics
Machine learning can diagnose diseases based on the amount of different RNA present in patient samples. However, this requires costly computers, equipment, and trained operators. DNA-based computers are a cheap and accessible alternative for reading and processing patient RNA. Currently, DNA computers are limited in processing power due to “leak”, interference from two interacting signals. By designing DNA computers that integrate with microfluidic devices, we can overcome leak for powerful point-of-care diagnostics.
Here we introduce DNA-based biomolecular neurons for low-cost machine-learning-capable DNA computers. These neurons assemble together to form in vitro neural networks. First, we use enzymes to create highly pure DNA components to reduce leak. Second, we immobilize the DNA strands onto magnetic beads for modular assembly. DNA by-products from neuron assembly and computation can then be removed from solution using magnets, further reducing leak. This immobilization also allows us to synchronize computation across each layer of the neural network. Integration with microfluidics will replace manual magnetic separation with automated fluid exchange, allowing massively parallel computing for neural-network-on-a-chip.
We demonstrate proof-of-concept for biomolecular neurons and enzymatic synthesis through modular primer mix-and-match and nucleic acid hybridization. We show synthesis of 6 unique species of neurons with input-specific actuation and high purity, requiring only 2 hours of preparation from start to finish. Our neurons actuate quickly, in less than 15 minutes, with 8-fold signal-to-background ratio, and are capable of making synaptic connections with up to 10 other neurons. Furthermore, we demonstrate self-assembly of these biomolecular neurons into computational layers, and show basic neural network features, such as neuron cascading, fan-in and fan-out motifs.
Our proposed design enables computer-free neural networks for complex decision making in portable diagnostics, autonomous lab-on-a-chips, compact data storage, and development of other in vitro neural network architectures. We plan for near future integration with microfluidic devices to bring further programmability and ease of operation to DNA-based computing.
Supervisor Name: Leo Chou
Year of Study: 4
Program of Study: PhD
Zoom link: https://us02web.zoom.us/j/89610372821?pwd=azd4SCtYVWtreVovaGNPV1c2NGY2Zz09
Meeting ID: 896 1037 2821
Powered by Calendly.com