>as an existential threat to his career and/or life's work.
Is it though? DeepMind stood on the shoulders of the giants: they made use of decades of biological research and wet lab experiments. There's much more to academic research than predictive data science, which is honestly not meant to be their expertise at all. It is exactly their research that enabled DeepMind to reduce the problem to a solvable deep learning problem, and it is still their research that can best leverage the results of the model. I think there's an important distinction between properly understanding and pursuing the science behind the problem and taking data and fitting an already formulated problem through a deep network.
While waiting for the paper, it seems to me "fitting an already formulated problem through a deep network" is not at all what AlphaFold did. Its main contribution is formulation of problem to be solved by a deep network.
What do you mean? Protein folding was already formulated as a deep learning problem in prior research. DeepMind used several engineering tricks that they previously used in their other DL work.
I am actually surprised at the way it was done, though. I would not have done it that way... I would have set up a 3 dimensional convolutional net and done training on time reversed melting simulation transitions.
Is it though? DeepMind stood on the shoulders of the giants: they made use of decades of biological research and wet lab experiments. There's much more to academic research than predictive data science, which is honestly not meant to be their expertise at all. It is exactly their research that enabled DeepMind to reduce the problem to a solvable deep learning problem, and it is still their research that can best leverage the results of the model. I think there's an important distinction between properly understanding and pursuing the science behind the problem and taking data and fitting an already formulated problem through a deep network.