Yes, the science advances, previously high-hanging fruits become low-hanging become high-hanging again [1], but the tooling also advances: we now have databases like OpenTargets which let us more easily evaluate potential drug targets. Failure is so much more than that, though: a program can fail after you've shown efficacy in animals, sometimes it just doesn't happen in the human subjects. Or you fail to find the right measurement (endpoint). A million ways to die.

[1] Gene editing is an example: impossible, then very possible, but now the blocker is public perception which in turn blocks investment.

I'm sure there's a treasure trove of already-built high-quality assets of Japanese trains.

China seems to make some strides for their own academics, thinking loudly about moving away from glam publications as career incentives: https://www.ft.com/content/64a811f1-b132-4211-8a8c-2252cf964...

It's supposedly about the worry of leaking state secrets, but it will have positive outcomes apart from that.

https://www.naturemetrics.com/species-detection

https://www.ednacollab.org/industry/

https://wilderlab.co/

These companies focus on environmental DNA - some are more on the level of local government monitoring, some are for private customers.

https://bufferbuffer.com/using-personas-in-technical-writing...

>In user experience (UX) design, personas are fictional characters representing the different types of users who might use a product or service. These personas are based on user research and are designed to help designers and other stakeholders understand the needs, goals, and behaviors of the different types of users.

The HN/YC crowd generally has software brain: https://www.theverge.com/podcast/917029/software-brain-ai-ba..., "when you see the whole world as a series of databases that can be controlled with the structured language of software code". Biology doesn't work like that most of the time, it's squishy and weird and unpredictable, and the models we have of biology (including genomics!) are faulty at best, misleading at worst. I've supervised PhD-students and it takes some time for people's brains to be comfortable with that squishiness, that random behaviour, that 'putting A into the system only rarely produces B and we don't really know why but we do it anyway' view of the world. Software engineers struggle, even abhor that kind of world, which is why you rarely see them being interested in it; and if they work in it, outcomes are sometimes amazing and Nobel Prize worthy, more often nonsense that silently disappears.

Modeling protein-protein binding is still a massively unsolved problem, mainly because we don't really have the data. Alphafold2 was great but didn't actually 'solve' protein-folding as all input data is from single 'state' X-ray crystallography of the proteins, not 'really' how these proteins behave in the wild. So it's still very, very had to predict what binds to what, which of course is a multi-billion-dollar industry.

I work in a pharma-field and I wish we could easily design molecular binders. We still spend millions every year finding targets that could 'smuggle' our drugs into cells.

Some other players in this field are Boltz Lab and Isomorphic Labs (the Alphafold Google spinoff led by Hasabi). None of them can predict anything complex or 'big', everything is peptide-level. OP's work is another step towards something better.

The most interesting part in the preprint is that they find no matches for their designed binders in the world-write protein database. An open question with protein-designers is whether they just regurgitate training material, which is far easier to test with English-language models.

Claude asked me three questions and then wrote a beautiful Python implementation that queries the map and spits out a table of adjusted probabilities. Felt immensely powerful - I can do this 'on my own' now, I don't need to wait to find the right people or learn the right thing first.