M.Sc Thesis

M.Sc StudentAbu Sinni Raghd
SubjectImplementing Memory in Genetic Circuits using Hopfield
Neural Networks
DepartmentDepartment of Biomedical Engineering
Supervisor ASSOCIATE PROF. Ramez Daniel
Full Thesis textFull thesis text - English Version


The field of synthetic biology aims at constructing synthetic gene circuits for the purposes of achieving new biotechnologies or advancing the research of natural systems. Many of these circuits are desired to have computational capabilities. For that, Conferring memory to genetic circuits is of paramount interest in this field. Memory enables achieving sustained responses from transient signals, recording of environmental signals and cellular responses and coordinating cellular events. From a computational point of view memory enables the construction of finite state machines and the implementation of sequential logic.

In this work we explore the use of neural network architectures, mainly Hopfield networks, in genetic and metabolic circuits, in order to confer associative memory to living cells. Neural networks are excellent candidates for cellular computation; they lie in a midpoint between the analog and digital paradigms, and attain advantages of both. They are characterized with low energy consumption, low part count, fast parallel processing and good noise tolerance.

To that end, we use the perceptgene developed by Daniel et.al [1]. It is an artificial neuron model inspired from gene regulation kinetics; It transforms the biological reaction to neural like nodes comprising multi weighted  input integration and thresholded activation. In this work, we Further extend the perceptgene model to fit other biochemical reactions, mainly phosphorylation systems and protein sequestration. Furthermore, we perform theoretical analysis to demonstrate that Hopfield networks are compatible with the new perceptgene model. More specifically we devised an appropriate energy function (Lyapunov function) and proved that spontaneous evolution of a perceptgene-Hopfield network minimizes this energy. We also conducted simulations that demonstrates the system’s ability to store multiple states, and discuss possible applications.