The Design-Build-Test-Learn cycle of Synthetic Biology
We are entering the era of Synthetic Biology. In this paradigm, we aim to industrialise biotechnology. This means that we embrace engineering principles along with those found in nature. The sizes of modern molecular datasets such as protein sequences and their structures are immense, and so is the number of possibilities for experimentation. This requires machine learning techniques and automation to navigate this biological landscape. Yet, human intuition, knowledge and science activities will continue to be the centre of directing scientific endeavours, as humans are the driving force behind bringing the changes that we want to see in the world. Synthexo implements Design-Build-Test-Learn cycles that embraces engineering and AI principles, and lets you direct your experiments. An example of such a cycle in a directed evolution experiment might look like this:
​
-
In the first round, use few-shot predictions (not AI training) and rational design (using human insights) to prioritise a series of backbone mutant proteins, and/or a list of mutant loci for saturation mutagenesis (depending on assay, throughput, costs).
-
​In the second round, a new AI architecture is built that can incorporate assay results from the first round along with PLMs and structural inputs, predicting assay outputs, and learning from results. This builds a fitness landscape in protein sequence space. We predict the next round of mutations.
- ​In the third round and onward, decisions will incorporate information from assays and human insights, and facilitated by navigation of the fitness landscape.
-
Depending on needs, larger and more complex structural changes could be attempted, using other AI methods, e.g. Diffusion/Flow matching to predict m​ajor changes to the backbone for IP purposes.