Towards a fully automated algorithm driven platform for biosystems design.

Area of Science:

• Synthetic biology
• Automation engineering
• Machine learning in biology

Background:

• The design, build, test, and learn (DBTL) cycle is crucial for biosystems design but faces challenges like high experimental costs, variability, and data analysis limitations.
• Traditional methods often miss key insights due to biases and the sheer scale of potential biological variants.

Purpose of the Study:

• To develop and demonstrate a fully-automated robotic platform integrating machine learning to optimize the DBTL cycle in biosystems design.
• To overcome the limitations of traditional experimental approaches in terms of cost, efficiency, and insight generation.

Main Methods:

• An integrated robotic system, BioAutomata, was coupled with machine learning algorithms to automate the DBTL process.
• A paired predictive model and Bayesian algorithm were employed to intelligently select experiments for optimization.
• The platform was validated by optimizing the lycopene biosynthetic pathway.

Main Results:

• BioAutomata successfully automated the DBTL cycle for biosystems design.
• The platform evaluated less than 1% of possible variants but outperformed random screening by 77% in optimizing the lycopene pathway.
• The system demonstrated effectiveness in black-box optimization problems characterized by expensive, noisy experiments and limited prior mechanistic knowledge.

Conclusions:

• Fully-automated robotic platforms like BioAutomata, powered by machine learning, can significantly enhance the efficiency and success rate of biosystems design.
• This approach offers a powerful solution for complex optimization challenges in synthetic biology, particularly when dealing with high-dimensional and noisy experimental spaces.