Living Biosensors at the Intersection of Genetic Circuits and Machine Learning
Synthetic Biology Engineering Evolution Design SEED
2021
2021 Synthetic Biology: Engineering, Evolution & Design (SEED)
Poster Session
Poster Presenters - Accepted
Advances in synthetic biology tools pave the way for engineering cells as biosensors which
take part in environmental monitoring, disease diagnostics and therapy, and bioproduction of
valuable compounds. To date, remarkable successful living biosensors have been
demonstrated in laboratory allowing cost-effective, user-friendly, renewable, and portable
prototyping for field deployment. Yet, living biosensors may suffer from certain challenges
such as leakage, specificity, sensitivity, multiplexing, and long response time. Many control
elements have been designed to solve most of the challenges. However, the more control
element in a circuit the more cellular burden, hence, lagging in sensor response has
resulted. In the last decade, synthetic biology collaborates with other disciplines (e.g.,
machine learning) to solve issues regarding to circuit design. In our study, we utilized a
neural network-based architecture to decrease the response time of a complex genetic
circuit which is engineered for gold ion detection. Thanks to Long-Short Term-Memory
(LSTM)-based networks, we were able to decrease the ON/OFF status of the sensor to 30
min with 78% accuracy and over 98% in 3 h. Next, we demonstrated that the network can
also classify the sensor output (raw fluorescence data) into pre-defined gold concentration
groups with 82% precision in 3 h. We envisage that this approach would be applicable to a
multiple living sensors with complex gene circuits which suffers from long response time.
This study has been supported by TUBITAK Grant No 114Z653 and 118S398.
take part in environmental monitoring, disease diagnostics and therapy, and bioproduction of
valuable compounds. To date, remarkable successful living biosensors have been
demonstrated in laboratory allowing cost-effective, user-friendly, renewable, and portable
prototyping for field deployment. Yet, living biosensors may suffer from certain challenges
such as leakage, specificity, sensitivity, multiplexing, and long response time. Many control
elements have been designed to solve most of the challenges. However, the more control
element in a circuit the more cellular burden, hence, lagging in sensor response has
resulted. In the last decade, synthetic biology collaborates with other disciplines (e.g.,
machine learning) to solve issues regarding to circuit design. In our study, we utilized a
neural network-based architecture to decrease the response time of a complex genetic
circuit which is engineered for gold ion detection. Thanks to Long-Short Term-Memory
(LSTM)-based networks, we were able to decrease the ON/OFF status of the sensor to 30
min with 78% accuracy and over 98% in 3 h. Next, we demonstrated that the network can
also classify the sensor output (raw fluorescence data) into pre-defined gold concentration
groups with 82% precision in 3 h. We envisage that this approach would be applicable to a
multiple living sensors with complex gene circuits which suffers from long response time.
This study has been supported by TUBITAK Grant No 114Z653 and 118S398.