The purpose of this project is to develop and program a self-folded computationally driven DNA robot that can facilitate a pathway across the blood-brain barrier and sense and respond to its environment, performing computation similar to a silicon-based computer. This system is based off of a neuron-like configuration from a base of an AND gate, and when connected forms a Hopfield neural network that is able to classify neuron cell types (astrocyte, neuron, OPC, microglia) based off of the concentration of RNA expressed from corresponding genes in the cells. In this way, it supersedes current drug-delivery methods in many aspects, as the accuracy in drug release is increased. Computation within a living system for the purpose of drug delivery could provide for much more accurate and reproducible drug-delivery systems, especially for CNS (central nervous system) disorders due to the acceptance of the blood-brain barrier to DNA.
A DNA robot was designed successfully with a computational lock system that after training with RNA sequence expression data of four different types of neuron cells most often affected by neurodegenerative diseases, classified four types of brain cells each with an accuracy of over 90%. It is important to notice the importance of the design for current treatments, and the ideas it presents towards intelligent medicine that could perform computation. As a first step towards such drug delivery and methodology, it provides a very accurate and known procedure for drug delivery. Further investigation determines physical fabrication of the robot and computational scalability.
The question investigated in this project is:
Can complex computation be replicated by a specially designed DNA nanorobot for computationally accurate drug-delivery by a sense-response system, similar to a macro-sized robot?
The purpose of this project is to develop and program a self-folded computationally driven DNA robot that can facilitate a pathway across the blood-brain barrier and sense and respond to its environment, performing computation similar to a silicon-based computer.
- supersedes current drug-delivery methods in many aspects as programmable computation within biological systems such as DNA can lead to high specificity and therapeutics that respond to their environment, important for current challenges and progress in reforming chemotherapeutics and the race towards improved accuracy and optimization in neurodegenerative drugs
- blood-brain barrier’s acceptance to DNA makes DNA a much more robust and less invasive vehicle option for drug delivery to the brain
The engineering goals are as follows:
(1) program and test the accuracy of this DNA robot’s classification of brain cell targets (and thus DNA’s computational abilities as a determinant of accurate drug delivery) by engineering a computational lock system on the robot that senses, processes information, and responds to the environment autonomously and
(2) design its carrier structure as the rest of the model that encompasses this lock system and facilitates its response (drug release or retainment) using DNA origami and brick principles.
For the experiment, the hypothesis and expected results are of a greater than 80% computational drug delivery accuracy.
Self-folding DNA Nanostructures
1. DNA origami to make intricate shapes . Marked a turning point in DNA nanotechnology, enabling unprecedented control over designing molecular structures.
2. “Brick” method provides a more robust and reproducible self-assembly model. 
3. Current applications of DNA origami designs:
- Self-assembling, self-destructing drug delivery vessels; researchers at Harvard’s Wyss Institute led by Shawn Douglas: DNA robots for targeting leukemia and lymphoma with some success.
- Cancer therapeutic in response to a molecular trigger (nucleolin) in vivo: DNA robot successfully circulated and released its payloads to cancerous cells simulated in mice 
- Cargo sorting DNA robot, which was capable of moving over the surface of a DNA origami sheet and sort molecular cargoes. 
1. Lulu Qian and Erik Winfree of Caltech: used DNA strand displacement cascades to construct a circuit composed of gates that could compute the square root of any number.  
2. Premise of DNA logic gates: more of a given input molecule, greater chances that it will displace the gate molecule on the gate, (can weighted and limited with a threshold)  - Logic AND and OR gates can be created with the DNA molecules. AND gate: output will reach high levels when both inputs are present at certain levels, which can be indicated and limited via a threshold (1). Otherwise, it will remain on the strand, indicating an output of (0).  - Include combinatorial problems, tic-tac-toe games, and development of Turing machines. 
3. First neural network composed of DNA developed by Lulu Qian: Hopfield neural network from DNA to answer four questions relative to two answers given to the computer. 
The Blood-Brain Barrier and Challenges in Delivery of Neurodegenerative Therapeutic Treatments
- Excludes ~100% of large-molecule neurotherapeutics and more than 98% of all small molecule drugs 
- Drug not in a therapeutically relevant concentration if crosses, binding of the drug molecule to other protein = drug ineffective (therapeutically/crossing barrier), enzymes in the brain tissue could render drug inactive. 
Please see bibliography section for all citations.
Design with DNA nanostructures began with creation of sheets of DNA that mimic certain predictated and engineering patterns. This self-assembly structure model was furthered by Peng Win's work work for DNA brick structures. Computation with DNA has been primarily investigated in great detail by Lulu Qian of Caltech, and her research has demonstrated that neural network (machine learning) computation is possible with DNA structures, as well that certain structures can be programmed for specific functions, such as a cargo-sorting robot. Drug-delivery past the blood brain barrier, a highly selective semipermeable membrane barrier, is difficult due to the selectivity and chemical properties of the barrier. However, DNA is a viable solution as a cargo vessel across the blood-brain barrier. Computationally accurate drug-delivery specific to cell-type is crucial in reforming current treatment.
Basis for DNA computation
DNA Origami and Brick Structures (Paul Rothemund, Peng Yin)
Layout of first design of DNA Origami (Paul Rothemund)
Current Neurodegenerative Treatment
- Carrier structure Self-assembled DNA nanostructure via “block” design Robust modeling
1. DNA sequences and positions
2. Reduced base pairing posterior end - spring mechanism 3. Lock anterior end, keeps closed - Output strand (<2 3>) released (1) on lock = tension of spring breaks remaining gate strand (<1* 2* 3* 4*>) = opens and releases payload
- Computation (DNA Strand Displacement) tool: rapid prototyping and analysis of computational devices
- Used to engineer and test unique gate configuration
1. Code for DNA AND logic gate written (inputs and output determination)
2. Thresholds set for gate (FPKM)
3. Test with mRNA expression levels and nucleotides
4. Products of reactions as determination of computation (1,0) analyzed
Pictured above: Developed Visual DSD code and schematic
Pictured above: Primary AND gate DNA strand, and Example expression for neuron threshold test with concentrations of each strand (nM)
Relative expression levels of RNA-seq in brain cells
1. Two most expressed mRNA sequences of each cell type
2. FPKM (expression level): threshold
3. mRNA sequence: toeholds designed to starting sequences
The experiment and testing process was ensured to be fair by cross-validating the computational outcomes. Therefore, by testing the model within itself (testing each neuron outcome against the other) it was ensured that all testing outcomes were previously unknown by the robot.
The experiment took place in the student's home, and software used was the NanoBricks software (design), Visual DSD (computational programming for DNA molecules) and the Brain RNA-Seq database for RNA sequences.
Methods in design:
The structure was designed in the NanoBricks software by specifying DNA connections within the structure. Specifically, certain DNA oligos were designated at certain positions for the structure, and the self-assembly method was simulated for the exact design. A reduced base pair hinge was created on the side in order to provide the tension for the opening of the structure to release the therapeutic.
Methods in computation:
A script was developed for the AND gate mechanism of the robot (both inputs must be present for the output strand to be released) within the language IDE. Then, thresholds were set within the code for each type of cell correesponding to the gate of a certain cell (neuron, astrocyte, OPC, microglia). These thresholds were tested in thousand iterations per cell, and then were mapped over time. True matches were recorded, as well as the opening of the robot to incorrectly designated cells.
Therefore, the integration of design and computation lay primarily within the mechanism for computation and testing/iteration.
Within the Visual DSD software exists two modules: one for code and one for DNA circuitry. The code written is for an AND gate, in which the inputs are strands < 1 2 > and < 3 4 >. These strands compete with the base strand <2 3 > attached via Watson-Crick pairing to <1* 2* 3* 4*>. In the code module, the variables N1, N2, and N denote the concentrations of <1 2 >, <3 4> and <2 3>, respectively. For example, the two genes most abundant in the neuron were tested as inputs first. These two genes, Vstm21 and Clstn2, have expression levels of 63.710 and 63.546 FPKM respectively, and their mean becomes the threshold. The DNA robot remains closed if gene expressed in any other cells attach to the gate, because their concentrations are not high enough to displace the output strand (overcome the threshold). Percent of the output strand that is released in the correct and incorrect cells was used to calculate accuracy, as well as SNR (signal to noise ratio). A large amount of original output strand released in correct thresholded gate indicates that the robot opened (lock broke and drug released) at correct cell. Percentages were calculated from nM of output strand released/nM of original output strand. SNR (signal-to-noise ratios) were reported as an indicator of the level of a desired signal to the level of background noise, thus displaying how well the robot pinpoints its classification. A high SNR value means that the robot identified the cell as not the target cell, and remained closed. Thus, accuracy can be seen as displacement and opening to a correct cell, as well as remaining closed to an incorrect cell (both are “correct” responses). After training the lock system for thousand iterations, the robot was accurate in its release over 90% in each cell (99.11% - Neuron; 97.81% - Astrocyte; 91.00% - OPC; 99.11% - Microglia). In addition, it remained closed for a majority of every incorrect cell (ex. Gate 1 [Neuron] - only released 12.96% output strand in Astrocyte, 30.66% in OPC, and 12.96% Microglia). It is important to note that Gates 2 and 3 (Astrocyte and OPC) provided outputs similar when classifying a neuron as the incorrect cell(~ 80%). The expression levels of these cells may be similar when tested by the gates, indicating that future work would aim to reduce these similarities or use different mRNA expressions as bases for this instance. Collectively, these data indicate that the utilization of DNA as a computation medium, especially for the application of classifying drug targets, displays high specificity and accuracy, and although error is present, it is readily quantifiable and reducible.
Full analysis: https://docs.google.com/document/d/1mHqZBybQ1_5g4kBmVIwckHEEWlh6E9FkYZeN3iHPVVk/edit?usp=sharing
A DNA robot was designed and developed successfully, with a computational lock system that after training with RNA-seq expression data of four different types of neuron cells most often affected by neurodegenerative diseases, classified four types of brain cells each with an accuracy of over 90%. In addition, the use of DNA for both the structure and computation, which consists of predictable AND gates, shows the robustness of DNA as an engineering structure (design and computational). In the future, methods to scale up the architecture and computation for greater accuracy within the system would be investigated. In addition, the next steps towards a nanorobot that could be incorporated into the human body and successfully perform computation would be taken, whether it is faster implementation, a reproducible method of isolation, or another method to elaborate on these designs. Utilization of DNA as a computation medium, especially for the application of classifying drug targets, displays high specificity and accuracy and thus could be very novel and revolutionary as a solution towards specific drug delivery.
Although the experiment demonstrates that there exists limitatitations in the accuracy of biologically based computation, it also shows that there exists potential for scalability and possibly the development of computational mechanisms to acheive the accuracy of silicon-based computers. Although the experiment was successful, it is evident that future approaches can be done for a greater emphasis on machine learning mechanisms and pattern recognition.
It is important to notice the importance of the design for current treatments, and the ideas it presents towards intelligent medicine that can perform computation. As a first step towards such drug delivery and methodology, it provides a very accurate and known procedure for drug delivery. In methods of cell recognition for increased accuracy, mistakes are often made, and these mistakes are often times difficult to predict, because the parameters are unknown. Ideas of computational medicine have the potential to revolutionize medicine and how researchers approach drug-delivery systems. With a DNA nanorobot that can sense and respond to its environment, further investigations could lead to scalable architecture with an infinite amount of functions both in and out of the body.
Specifically, further questions include the robustness and reproducibility of biologically based computation in vitro and in vivo. Could a fabricated DNA robot be similarly programmed and then act on decisions similarly to how it had done in the project? Once experimented in vivo, would this model have the potential and computational accuracy for significant accuracy in drug delivery for neurodegenerative diseases? If further scaled, what further impacts and applications could very accurate and specifically engineering molecular robots have for current challenges such as drug delivery, medicine, or other micromolecular factors in the human body?
My name is Sofiya Lysenko, and I love the expansive and unlimited potential that arises from the intersections of biology and computer science. Specifically, I have a passion for computer science, and am fascinated by how we can computationally program or instill computation into living things or components. I believe the future of computation will be biological.
From a young age, I discovered passions in both computer science in biology in robotics camps and science fairs. In elementary school, I investigated topics such as immunology and nanotechnology. In 6th grade, I built a homemade lie detector with Arduino, and after investigating the effect of the golden ratio on flight in the 7th grade, I built my own homemade gel electorphoresis device at home in the 8th grade and extracted protein from onion, bannana, and kiwi. Then, I realized that Zika virus mutations could be analyzed by a gel electrophoresis device and machine learning algorithm. After receiving the 4th place national prize at the ProjectCSGirls competition, I discovered the Cello programming language and programmed bacteria to express drug-conjugated antibodies for more efficent and personalized cancer treatment. I plan to continue in research as PhD and then principal investigator, investing towards my own laboratory/companies.
I hope that winning would advocate and push molecular robotics and ideas within biological computation to a greater forefront! The prizes would allow me to further the impact of this emerging field and change my life in their inspiration and meaning for the furtherment of the field's potential.
I was mentored by Jocelyn Kishi (email@example.com) a postdoctoral student at the Wyss Institute of Biologically Inspired Engineering at Harvard University. The mentorship was conducted remotely, and all experimentation was done at home.
Thank you to Jocelyn Kishi, a graduate student at the Wyss Institute at Harvard University, for her mentorship and guidance in this project as well as Dr. William Shih (Wyss Institute) and Dr. Lulu Qian (Caltech) for their willingness to communicate and answer questions throughout the process.
Thank you to Mr. William Anderson, Ms. Doretta Agostine, Mr. Ryan Cragle, and Dr. Matthew Hartwell at the Abington Senior High School, Emily Arturo and Dr. Amanda Purdy at the Fox Chase Cancer Center, Erika DeBenedictis, a graduate student at the Massachusetts Institute of Technology, and Dr. Michelle Johnson’s Rehabilitation Robotics Laboratory students at the University of Pennsylvania and my lab mentor Wilson Torres for their insightful and helpful feedback and advice on the project.
 Rothemund, Paul W. K. “Folding DNA to create nanoscale shapes and patterns.” Nature News, Nature Publishing Group, 16 Mar. 2006, www.nature.com/articles/nature04586.
“Ten Years of DNA Origami | Caltech.” The California Institute of Technology, www.caltech.edu/news/ten-years-dna-origami-50125.
“DNA origami.” Wikipedia, Wikimedia Foundation, 5 Mar. 2018, en.wikipedia.org/wiki/DNA_origami.
Ke, Yonggang, et al. “Three-Dimensional Structures Self-Assembled from DNA Bricks.” Science, American Association for the Advancement of Science, 30 Nov. 2012, science.sciencemag.org/content/338/6111/1177.
“DNA origami could allow for 'autonomous' delivery.” FiercePharma, 15 May 2012, www.fiercepharma.com/r-d/dna-origami-could-allow-for-autonomous-delivery.
 Li, Suping, et al. “A DNA nanorobot functions as a cancer therapeutic in response to a molecular trigger in vivo.” Nature News, Nature Publishing Group, 12 Feb. 2018, www.nature.com/articles/nbt.4071.
Thubagere, Anupama J., et al. “A cargo-Sorting DNA robot.” Science, American Association for the Advancement of Science, 15 Sept. 2017, science.sciencemag.org/content/357/6356/eaan6558.
“Blood–Brain Barrier.” Wikipedia, Wikimedia Foundation, 31 Mar. 2018, en.wikipedia.org/wiki/Blood–brain_barrier.
“Nanoparticles for Drug Delivery to the Brain.” Wikipedia, Wikimedia Foundation, 30 Mar. 2018, en.wikipedia.org/wiki/Nanoparticles_for_drug_delivery_to_the_brain.
“Nanoparticle-Mediated Brain Drug Delivery: Overcoming Blood–Brain Barrier to Treat Neurodegenerative Diseases.” Journal of Controlled Release, Elsevier, 18 May 2016, www.sciencedirect.com/science/article/pii/S0168365916303236.
“Drug Delivery to the Brain.” Wikipedia, Wikimedia Foundation, 30 Mar. 2018, en.wikipedia.org/wiki/Drug_delivery_to_the_brain.
Patel, M M, and B M Patel. “Crossing the Blood-Brain Barrier: Recent Advances in Drug Delivery to the Brain.” CNS Drugs., U.S. National Library of Medicine, Feb. 2017, www.ncbi.nlm.nih.gov/pubmed/28101766.
 “Delivering Genes Across the Blood-Brain Barrier | Caltech.” The California Institute of Technology, www.caltech.edu/news/delivering-genes-across-blood-brain-barrier-49679.
Qian, Lulu, and Erik Winfree. “A Simple DNA Gate Motif for Synthesizing Large-Scale Circuits.” Journal of The Royal Society Interface, The Royal Society, 4 Feb. 2011, rsif.royalsocietypublishing.org/content/early/2011/02/03/rsif.2010.0729.full.
Technica, Ars. “Synthetic Biologists Use DNA to Calculate Square Roots.” Wired, Conde Nast, 3 June 2017, www.wired.com/2011/06/dna-mathematics/.
“DNA Computing.” Wikipedia, Wikimedia Foundation, 31 Mar. 2018, en.wikipedia.org/wiki/DNA_computing.
Qian, Lulu, et al. “Neural Network Computation with DNA Strand Displacement Cascades.” Nature News, Nature Publishing Group, 20 July 2011, www.nature.com/articles/nature10262.
Lakin, Matthew R. et al. “Visual DSD: A Design and Analysis Tool for DNA Strand Displacement Systems.” Bioinformatics 27.22 (2011): 3211–3213. PMC. Web. 5 Mar. 2018.
“The Qian Lab.” The Qian Lab - Caltech - Research, www.qianlab.caltech.edu/research.html.