In his remarkable work and in his TED talk, author and inventor Steven Johnson noticed that good ideas are rarely the result of a single individual brooding over a complicated concept. Instead of being a shiny, single instance that arises suddenly out of nowhere, ideas somehow seem to be the result of networks that feed off a chaotic, rich, and diverse background.
This highlights the intrinsic value that cross-discipline approaches hold – recognized as well in the funding schemes of for example Horizon 2020 or the Melinda and Bill Gates foundation, which place special importance on having different fields interact with each other. Our recent work is a showcase for the fruitfulness of knocking on the door of your neighbors and join forces to tackle a project.
It started off with a simple idea firmly set in the comfort zone of the field that was first involved in this project, genetic engineering: The creation of a genetic circuit to control a pathway in an orthogonal fashion, realized by orthogonal sigma factors (see Step 1 in figure). In order to optimize pathways, different promoter strengths are indispensable. The search began: From orthogonal promoter libraries, colonies with reporter genes were tested for their expression strength. Here, the project still lied calmly in the realms of synthetic biology, following our beloved Design-Test-Build-Learn cycle in laborious, small steps.
Figure showing the development from a sigma factor toolbox for pathway optimization to a promoter strength prediction tool. Figure taken and slightly adapted with permission from M. Van Brempt.
As the work progressed, it became more and more obvious that cell sorting combined with next-generation sequencing could streamline the project, allow for the selection of discrete promoter strengths and make pathway optimization (Step 2 in figure) feasible. The side-effect of these techniques is the accumulation of data. A lot of data.
It so happened that a project with our neighboring group for Mathematical and Computational Modelling kicked off and that was when a new, shiny, and very promising question rose to the surface: Could we collaborate to elevate the research problem from constructing what we needed to optimize a pathway to making a predictive tool applicable to all sorts of questions (Step 3 in figure)? Can we combine the predictive power of machine learning and the tunability of microorganisms to control gene expression?
It turned out that it was indeed possible. The resulting forward engineering tool, built on convolutional neural networks and published on November 16, 2020, is able to predict promoter transcription initiation frequency of sigma factor-specific promoters and can provide orthogonal promoter sequences at a desired strength. We were able to validate such a library in vivo. In summary, the project wanted to demonstrate that we can find orthogonal promoters of specific strengths to optimize pathways. The project ended up demonstrating that it can be even more interesting to go beyond your research field and find the underlying bigger question.
Joining forces always means encountering curiosities and shared moments of bewilderment. Sure enough, we can look back and cherish plenty of funny moments of discovering what it really meant to work in the respective field of synthetic biology and machine learning. Like when the machine learning people asked if the strains were finally ready, unfamiliar with of how laborious and time-consuming cloning work can be. Or when the wet lab people pressed ‘run’ on their model for the first time and then stared at the screen, wondering how long they would have to wait for an answer from this so-called neural network.
A take-home message for us was certainly that sometimes, 2 + 2 = 5. As we look beyond our usual horizon and ask the same question to another person, obstacles can suddenly transform into an opportunity. Being curious about what possibilities might be hiding in our neighbor’s strange work might well lead to your next big project.