Fieldability, or implementation of synthetic biology constructs out of the laboratory, has the potential to help solve numerous global problems. However, currently, synthetic biology does not have the infrastructure to develop and implement fieldable systems. There are several roadblocks that prevent fieldablility from being a reality. One of these roadblocks stems from a lack of knowledge surrounding which chassis is ideal for a specific deployment environment. As numerous different factors impact bacterial survivability, it is difficult to predict where a specific chassis will survive. However, in order for fieldable genetic systems to function accurately, the chassis needs to be metabolically active and viable in the deployment environment. To help synthetic biologists designing fieldable systems, the 2022 William & Mary iGEM team developed a novel, 16S data-driven, model-based software program to predict the optimal bacterial chassis for an air, water, soil, or human gut microbiome environment.
Inspiration
From the beginning of our project, our team was interested in fieldability, or implementing synthetic biology constructs outside the laboratory. While synthetic biology has the potential to solve global problems, many solutions will require fieldable systems. These systems range from smart drugs employed in human intestines to biosensors that track pollution levels in a lake. Fieldability would allow synthetic biology to restore habitats, degrade plastic on-site, and even increase drought tolerance in plants. As we learned through our IHP, the potential for fieldable synthetic biology is endless.
When we started our project, we considered creating a genetic system to help bioremediation pollutants in the environment. However, as we conducted our literature review, we found that synthetic biology does not currently have the infrastructure for implementing fieldble genetic systems and that there are several barriers preventing fieldability from becoming a reality. Originally, our team planned to tackle several of these barriers through a wetlab project. However, after our IHP interview with Mr. Marken, the entire direction of our project changed. As we explained our old project idea to Mr. Marken, he stressed the importance of the software that we planned to develop and the value of computation to synthetic biology. After this interview, our team was motivated to develop a largely computational project, something that our team has never done before. In previous years, our team has structured itself into a “wetlab” team and a “computational” team. However, after this meeting, it was clear to us that every team member should be computational! In response to this realization, much of our education efforts this year have been dedicated to stressing the importance of computation to synthetic biology and biology as a whole.
Significance
Fieldability has the potential to solve numerous different global problems. However, there are several barriers to developing fieldable systems, and unfortunately, without addressing these roadblocks, fieldable synthetic biology cannot become a reality. Some of these barriers include ensuring safety, assessing output, and selecting an optimal chassis for specific out-of-laboratory environments. Within these different roadblocks, optimal chassis selection is the first obstacle that a researcher must overcome.
A chassis is the organism that hosts a genetic construct. Without this organism, the construct is unable to function. Chassis choice impacts every aspect of circuit design and implementation; therefore, to start designing fieldable systems, researchers must pick the proper chassis for their target environment. For the genetic construct to function correctly, the chassis must be viable and metabolically active in the deployment environment. Although bacterial survivability is crucial to fieldable circuit implementation, it is difficult to predict where a bacterial chassis will survive. There are many different factors that can influence their survival, ranging from abiotic factors, such as salinity and pH (Sridhar, 2022), to biotic factors, such as phage and competition with local microbes. These numerous different factors make it incredibly difficult to choose a chassis that will function optimally in a given environment. In order to develop fieldable systems, synthetic biology needs a method of predicting bacterial survival in different environments.
Computation has been a part of synthetic biology since the field’s inception and has played an important role in advancing the field. While there are several computational programs to assist researchers with designing circuits, currently, there is no software program to help researchers choose an optimal bacterial chassis for a specific deployment environment. According to Sidhar et al., “choosing the right chassis organism is just as important as choosing ideal genetic elements”. Therefore, software predicting the survival of chassis in different environments outside of the laboratory would be equally important to these design programs. Given the vast amount of 16S data available, our team believes that computational systems are the perfect method with which to predict bacterial survival and create this much-needed chassis prediction program.
Our Solution
The 2022 William and Mary iGEM team developed a 16S data-driven, model-based software tool to assist researchers in selecting a chassis for an air, water, soil, or human gut microbiome environment. In order to give the most complete description of bacteria survival, we employ a three-pronged approach, utilizing artificial intelligence, multivariate regression, and genome-scale metabolic models (GEMs). We implemented an independent prediction strategy using an Artificial Neural Network structure created in Tensorflow and Keras. Using a subset of relevant parameters as a 1D input layer, our training process optimizes the weights and biases in the network such that an output layer can accurately predict the relative abundance of certain species and genera in a given environment. Our final neural network is structured to consist of 5 fully connected dense layers between input and output nodes, which are optimized using the Adam optimization algorithm and a Mean Squared Error loss function.
In addition to analysis through a neural network, we also used our large data set to conduct a multivariate regression. Our regression includes high degree linear terms as well as logarithmic and interactive variables. By minimizing the mean squared error, we gathered interpretable, statistically relevant, and quantitative relationships between soil parameters and individual species relative abundance.
Lastly, we incorporated Genome-Scale Metabolic Models to simulate the metabolic success of individual species in suboptimal environments. GEMs are stoichiometric matrices that map metabolites to all reactions in a cell. We specifically used flux balance analysis (FBA) to isolate the reactions necessary for metabolic processes and optimize biomass as an objective function. Using linear programming, we found the rates at which metabolic compounds are converted into biomass components such as nucleic acids, proteins, and lipids.
Our software toolkit consists of several different packages that enable researchers to determine the optimal bacterial chassis for a specific environment. This software package consists of separate programs encoding a neural network (NN), multivariate linear regression, and GEM models. When environmental conditions (such as temperature) are input into our NN and linear regression, these tools are able to predict the relative abundances of species and genera in that specific environment. By using species-specific GEM models, our software is able to predict the growth rate of potential chassis, with available GEMs, in that same environment. Each individual program will output its interpretation of the optimal bacterial chassis for that environment. By comparing these outputs, researchers can determine which species to use as a chassis.
To learn more about the components of our software, please see our design page.
To view our software toolkit, please visit the Gitlab: chassEASE.
Works Cited
Leggieri, P., Liu, Y., Hayes, M., Connors, B., Seppälä, S., O’Malley, M. & Venturelli, O. (2021). Integrating Systems and Synthetic Biology to Understand and Engineer Microbiomes. Annual Review of Biomedical Engineering. 23, 169-201. https://doi.org/10.1146/annurev-bioeng-082120-022836
Sridhar, S., Ajo-Franklin, C., & Masiello, C. (2022). A Framework for the Systematic Selection of Biosensor Chassis for Environmental Synthetic Biology. ACS Synthetic Biology. 11(9), 2909-2916. https://doi.org/10.1021/acssynbio.2c00079