One thing I really admire about isomorphic and DeepMind in general is that they really try and encode human ingenuity into their models. The only reason that AlphaFold worked wasn't because they had better hardware or more money to train models - it's because they used every bit of information they had access to.

The same goes for AlphaGenome - it only takes a few hours to train the model, but it's freaking amazing what it can predict.