The software is useful because it allows you to have some idea of the function of each component prior to their assembly. Further, at this point, the software is designed for R&D far far upstream of being put directly used as a therapeutic. Finally, if a design causes some problem, we want to know that - we allow that undesirable property to be discovered in the laboratory, annotated, and communicated so that others know to not use such designs.
Is it actually possible to construct proteins from functional components like we design digital logic and software? Don't the parts interact in highly complex ways that take a lot of compute time to solve?
If you're trying to compute from first principles (atom by atom), yes, exactly - that's a very hard problem. And if you're rearranging entire protein domains rather than point mutations - that's an intractable problem. And it is precisely the problem we are trying to solve - how can you have any idea what the function of a novel protein will be when you are swapping around hundreds of amino acids at a time?
We've taken a different approach. A 'small data' approach, where we manually, and semi-automatedly ingest hard-earned empirical data about each domain, and make it computationally accessible. We give heuristics and logic to a biologist's intuition about 'this protein never seems to work when it's at the N-terminus', rather than trying to compute why it doesn't work by simulating the movements of 100,000 atoms over 10ns. And though many (generally synergistic) interactions won't be predictable in this manner, many will be. We help get you to those designs likely to work as quickly as possible. And the more empirical data ingested, the faster we think we can get you there. Further, even if it's not 100% effective at predicting the single best design, enriching the search space of a large (and expensive) screen is itself valuable.
Life already mixes & matches DNA to create new functional all the time, we're just making that same process explicit and accessible. It turns out the function of any given protein domain is actually reasonably robust to being split up and rearranged. Especially when the goal is 'some function', or 'good enough', and you're not messing with or trying to tune a protein absolutely required for life.
