I put a slightly longer version of these thoughts on msuskiewicz.github.io/blog/

P.S. When another revolution, cryoEM resolution one, took off in 2010s, what contributed a lot to progress alongside new tech was the old fashioned mol biol and biochem skills (sample preparation, optimisation). What fuels the AI revolution in maths seems to be the human ability to do maths by hand.

Lastly, if all this AI business is really about human-collected and -created data, and whether methods and insights extracted from them extend to new cases -- then it is clear that at some point one also needs new data, and every bit of new data is precious in itself and by adding to training sets.

Not to mention that currently most attempts to prove complex maths problems with AI yield incorrect results: you need ways to distinguish the rare correct ones, and understand their novelty (or unexpected triviality) etc. - something only an expert can do.

With time limitations become apparent - but the tool does have a profound impact. What always remains most valuable is human attention: setting what is interesting, posing the question, critically evaluating/crosschecking, understanding and explaining to others the result, moving beyond limitations.

applying known methods to new problems), although some higher-level 'rules' will probably be locally found in some types of problems. But perhaps more importantly AF revolution teaches us that real progress can be seen with naked eyes and doesn't need the hype that surrounds it at the beginning.

(e.g. antibody:antigen or protein:nucleic acid interactions seem currently to accurately work only if there is some homology to things already in the PDB). In fact, the use of AI for maths seems to mainly work through 'homology' so far (finding existing but overlooked solutions in old literature,

But it does change how things are done, offering ways of doing things faster or in a more high-throughput way, especially in the realms where it managed to infer some higher-level rules (standard protein modelling) and in other realms to the extent that queries resemble bits of training set

I used to love maths but know little of its professional academic side, so whatever I think of it is surely very naive. But the analogy to AlphaFold is perhaps useful, and we as structural biologists can offer an insight. AF has not really invalidated any older approaches, nor is it infallible.

accumulate individually a brighter focus - but if these are the same loci then I understand they might be similarly intense with pol II. In any case, thank you for the explanation! Lovely paper btw

I meant in comment 2), and I didntt understand immediately these are the same loci on two chromatids rather than different proximal genes (but if it is the same FISH probe then I see that these would be the same loci). If these were different genes, I was wondering why such proximal genes would each

reason for such proximal genes in particular to each individually get a brighter cluster? It might be that I'm confused. In any case - I thank you very much for your clarification. Lovely study btw