Assuming we had millions of 3D-scans of different kinds of animals and their full DNA sequence. Wouldn't it be interesting to try to use generative ML methods to generate a dinosaur DNA based on a 3D model?
There’s just no way this would work, the DNA codes organs, proteins and all of the complex molecules and molecular interactions, there’s no way a generative model is going to glean all that from 3D scans.
Could you normally have two viable lineages, each with the same DNA, and each with a different expression of that DNA which is stably conserved over many generations, purely due to womb environment, etc.? In other words, if you cloned a woolly mammoth using an elephant surrogate, might the great great great great grandchildren of this clone still have some characteristics that are due to having an elephant surrogate ancestor?
Or would there be a tendency to converge to a single stable expression due entirely to genetics?
It‘s obvious that in principle the answer could go either way, but I’m not sure whether that’s true in practice, with naturally occurring organisms and naturally occurring DNA sequences. For the sake of this question, one is also tempted to exclude post-natal “cultural” transmission, but it’s not clear that can be easily distinguished.
But I'm sure you can start there and find some interesting tangent that can very well turn out to be worth pursuing further.