Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Portland State University Portland State University
PDXScholar PDXScholar
Student Research Symposium Student Research Symposium 2018
May 2nd, 11:00 AM - 1:00 PM
Biochemical Reservoir Computing Biochemical Reservoir Computing
Hoang Nguyen Portland State University
Christof Teuscher Portland State University
Follow this and additional works at: https://pdxscholar.library.pdx.edu/studentsymposium
Part of the Computer Engineering Commons, and the Electrical and Computer Engineering Commons
Let us know how access to this document benefits you.
Nguyen, Hoang and Teuscher, Christof, "Biochemical Reservoir Computing" (2018). Student Research Symposium. 8. https://pdxscholar.library.pdx.edu/studentsymposium/2018/Poster/8
This Poster is brought to you for free and open access. It has been accepted for inclusion in Student Research Symposium by an authorized administrator of PDXScholar. Please contact us if we can make this document more accessible: [email protected].
Hoang Nguyen – [email protected]
Adviser: Dr. Christof Teuscher
Biochemical Reservoir Computing
Maseeh College of Engineering and Computer Science
Portland State University
BACKGROUND AND
MOTIVATION
METHODS
CONCLUSION
REFERENCES [1] A. Goudarzi, M. R. Lakin, D. Stefanovic, “DNA Reservoir Computing: A
Novel Molecular Computing Approach,” 2013.
[2] D. Angeli, “A Tutorial on Chemical Reaction Network Dynamics,” European
Journal of Control, vol. 3, no. 4, pp. 398–406, 2009.
[3] R. Rojas, “Neural Networks: A Systematic Introduction,” Springer, 1996
𝑑[𝑃1]
𝑑𝑡 = ℎ𝛽 𝑆1 𝐺1 − 𝑃3 −
𝑒
𝑉[𝑃1],
𝑑[𝑆1]
𝑑𝑡 = 𝑆1
𝑚 − ℎ𝛽 𝑆1 𝐺1 − 𝑃3 −𝑒
𝑉[𝑆1],
𝑑[𝑃2]
𝑑𝑡 = ℎ𝛽 𝑆2 𝐺2 − 𝑃1 −
𝑒
𝑉[𝑃2],
𝑑[𝑆2]
𝑑𝑡 = 𝑆2
𝑚 − ℎ𝛽 𝑆2 𝐺2 − 𝑃1 −𝑒
𝑉𝑆2 ,
𝑑[𝑃3]
𝑑𝑡 = ℎ𝛽 𝑆3 𝐺3 − 𝑃2 −
𝑒
𝑉[𝑃3],
𝑑[𝑆3]
𝑑𝑡 = 𝑆3
𝑚 − ℎ𝛽 𝑆3 𝐺3 − 𝑃2 −𝑒
𝑉𝑆3 ,
Reservoir computing is an emerging machine learning paradigm. Compared to traditional
feedforward neural networks, the reservoir can be unstructured and recurrent and only the
output layer is trained. Reservoirs can be built with various types of physical components,
yet, biochemical building blocks have not been widely used. This project focuses on
designing and testing a reservoir computer (RC) based on chemical reaction network
(CRN). We simulated high-level CRNs in MATLAB and their complex chemical
dynamics were observed over time. A CRN constructed by a network of coupled
deoxyribozyme oscillators was chosen for the final RC model. This project thus has the
potential to influence biomedical research and genetic disorder treatments.
RESULTS
The CRN dynamics inside the reservoir were mathematically model by a set of ordinary
differential equations (ODEs) shown in (1)
Key parameters:
[𝑃𝑖]: 𝑐𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑝𝑟𝑜𝑑𝑢𝑐𝑡 𝑚𝑜𝑙𝑒𝑐𝑢𝑙𝑒𝑠
[𝑆𝑖]: 𝑐𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑢𝑏𝑠𝑡𝑟𝑎𝑡𝑒 𝑚𝑜𝑙𝑒𝑐𝑢𝑙𝑒𝑠
[𝐺𝑖]: 𝑐𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑔𝑎𝑡𝑒 𝑚𝑜𝑙𝑒𝑐𝑢𝑙𝑒𝑠
𝑆𝑖𝑚: Influx rate of the substrate molecules
1 0 1 0 0 1
0 1 1 1 0 1
Hamming
Distance
In the readout layer of the reservoir computer, we implemented a single perceptron that read
random information from the reservoir and learn the correct Hamming distance between 2
input bitstreams. In our experiments, we feed the influx rates and the concentration at a
random time of each substrate molecule to the reservoir. For simplicity purpose, the basic
perceptron algorithm with sigmoid function was employed.
ACKNOWLEDGEMENT This material is based upon work supported by the National Science Foundation
under grant no. 1518833. The original RC model of this work was described in
“DNA Reservoir Computing: A Novel Molecular Computing Approach” by
Goudarzi et al. The authors acknowledge the support of the teuscher.:lab, the
Maseeh College of Engineering and Computer Science, and the University of
New Mexico.
Figure 2: Amplifying oscillator Figure 3: Attenuating oscillator
Figure 4: Single-input reservoir computer
Figure 5: Two-input reservoir computer
Random perturbations were introduced at different times to produce random oscillations of the product
concentrations. In this RC model, substrate(s) were perturbed in every τ seconds to introduce reservoir
inputs. Figure 4 illustrates single perturbations and figure 5 illustrates double perturbations.
In the lab, the system of ODEs (1) were solved and simulated with MATLAB ode23t
solver using Runge – Kutta method.
In order for the reservoir to start oscillate, the chemistry was unperturbed for 500 seconds, following
by a change in the influx rate of at least one substrate. With a single perturbation, this reservoir showed
dynamics of an amplifying and attenuating oscillator as shown in figure 2 and 3.
OBJECTIVE Design and build a reservoir computer (RC) model based on a
system of chemical reaction networks (CRNs) that has
computational and learning capabilities.
ANALYSIS
Here we show that a reservoir computer constructed by a system of coupled
deoxyribozyme oscillators is capable of performing computational and learning
tasks. From the experiments, the results show that such an RC can successfully
learn linearly separable patterns, with an average learning error of 0.0031. Further
research will be conducted to minimize the learning error and increase the
learning success rate. In addition, DNA strands will be feed into the RC and will
be interpreted as the binary bitstreams. This model can also be further designed to
detect pathogens and genetic mutations.
The computational abilities of this coupled deoxyribozyme RC model were tested
by learning the Hamming distance. 1000 different sets of 6-bit inputs were
applied to the reservoirs consecutively in 1000 experiments. Table 1 below
illustrates the average error of each bit on a sample simulation, together with their
targets and actual outputs.
Bit index # Target Actual output Average error
1 0 5.2396e-04 0.0028
2 1 0.9994 0.0025
3 0 2.1464e-10 0.0036
4 0 2.7559e-05 0.0029
5 1 0.9995 0.0034
6 0 6.4227e-04 0.0031
Table 1: Results of a sample Hamming distance learning performed by the
coupled deoxyribozyme oscillators
(1)
Figure 6: DNA strands with genetic information Source: https://en.wikipedia.org/wiki/DNA
Figure 1: An abstract reservoir computer with perceptron implemented