Ph.D Thesis

Ph.D StudentRizik Luna
SubjectSynthetic Neuromorphic Computing in Living Cells
DepartmentDepartment of Biomedical Engineering
Supervisor ASSOCIATE PROF. Ramez Daniel


The field of synthetic biology builds synthetic gene circuits in order to understand natural systems and to construct new biotechnologies. The dominant computational paradigm in synthetic biology is digital, like in electronics. The digital paradigm has enabled scientists to implement various applications in living cells, such as switches, counters, logic gates, memories, and edge detectors. Implementing digital circuits requires many genetic parts that are orthogonal and expressed at a high level. While significant progress has been made in creating large digital circuits with orthogonal synthetic parts1, digital circuit complexity is ultimately going to be limited by the fact that there are only 100s-1000s of transcriptional factors and 1000s-10,000s total genes per cell in most organisms. To address the limits of digital gene circuits, Daniel previously implemented an analog computational paradigm in living cells2. Analog computation computes with a continuous set of numbers and scales significantly more efficiently than digital computation. They implemented complex calculations such as addition and ratios with few transcription factors. However, like in electronics, analog circuits are subject to design issues such as noise, reliability, and loading effects that limit their scale. Consequently, a joint analog-digital paradigm, such as neuromorphic computing, is capable to smoothly handle synthetic gene circuits design challenges and to implement more complex functions.

Neuromorphic computing is an emerging engineering discipline that applies abstract models of neural systems to physical systems such as microelectronics3 or optics4 to build large-scale artificial intelligence systems. These artificial intelligence systems can efficiently solve complex perceptual tasks that were previously unique to living systems. Furthermore, the neuromorphic paradigm smoothly handles the design challenges and tradeoffs inherent to scaling complex digital networks, suggesting that neuromorphic computation may eventually replace the digital computational paradigm which has dominated electronics for more than 50 years. In contrast to electronics, neuromorphic computing has not yet been implemented in synthetic gene circuits in living cells.

In this study we demonstrate the first neuromorphic synthetic gene circuits. We developed a new fundamental model that maps neural networks to molecular networks using logarithmic transformation, to implement neuromorphic computing in living cells. These neuromorphic synthetic gene circuits are an important step forward because they enable living cells to execute fuzzy logic and synthetic computation that are impossible with digital and analog circuits, analogously to how artificial neural networks enable computer artificial intelligence. And, like computer artificial intelligence, we show that with backpropagation and supervision these neuromorphic synthetic gene circuits are trainable and can learn. We also show that neuromorphic synthetic gene circuits are scalable and can be composed with other types of synthetic gene circuits to create higher-order functions. Finally, we applied neuro-inspired data converters to encode analog cellular information by digital multi-bit decision. These data converters can be useful in many fields such as, therapeutic applications, food and fuel industry, and for detecting of pollutants and toxins by using only one inducer. This study marks an important step towards adaptive synthetic biology, which offers promise for emerging biotechnology and therapeutic applications that require complex and cognitive computations.