Isomorphic Labs: Reimagining drug discovery with an AI-first approach | Hacker News Reader

Isomorphic Labs: Reimagining drug discovery with an AI-first approach

isomorphiclabs.com

75 pointsXoS-49022 comments

> [...] there may be a common underlying structure between biology and information science - an isomorphic mapping between the two - hence the name of the company. Biology is likely far too complex and messy to ever be encapsulated as a simple set of neat mathematical equations.

...uses a neatly-defined mathematical term to name the company, then rejects the idea of neat mathematics in practice.

using arbitrary function approximators no less. everything can be isomorphic if you stir the linear algebra pot for long enough!

Do you disagree with that? I think the point they’re making is perfectly reasonable

See the poster's username, he/she is being a little too pedantic here :)

Funnily enough, once of my mathematically inclined friends has had this exact issue with biological problems. Every time he tried breaking it down into a neat solution, the response from the bio side would inevitably start with "but" or "however".

What Is Life? [0]

[0] https://en.wikipedia.org/wiki/What_Is_Life%3F

Was alalphafold2 that big of a landmark in application sense? I realize the protein folding problem is a hard and important one. But I didn't realize this result had immediate practical application. Will they just be consulting pharmas? Patienting proteins and such?

I think it could be Nobel Prize worthy. Protein’s structure often determines its effect as a catalyst. So to map the DNA to the 2nd order outcomes seems like it could be the missing ingredient to controlling the properties of cells.

Personally I’m hoping that someone smarter than me figures out how to displace existing catalysts like platinum and palladium. Seems like it could be a pretty penny and some positive environmental impact to boot.

Pt and Pd are pretty different from proteins in terms of the reactions they’re used to catalyze, but check out https://opencatalystproject.org/

This is wonderful. I’m going to see if there is anything I can build for this.

Protein catalysts almost always contain an inorganic atom. We will be able to make more efficient catalysts with better understanding of protein structure but the need for metals won't necessarily go away.

The protein catalysts can utilize more common metal ions instead of the rarer metals. Placing an iron atom in the right protein structure will affect the reaction probabilities.

Not really - it's just consistently better than everyone else, and obviously the best solution that anyone is likely to come up with for a while. But there were already plenty of examples of near-perfect de novo structure predictions from other groups, going back more than a decade.

Can someone explain to me why this should be an entirely new company (subsidiary) rather than folding DeepMind's capabilities into Verily and Calico? Are these different groups siloed from one another within Google?

Calico seems to be focused on combating aging.

I believe there is a lot of internal politics and dynamics playing here.

If you are Demis Hassabis and want to lead this new initiative:

- Why would you want to report to the CEO of another subsidiary instead of spinning-off another entity within the Alphabet conglomerate?

- You can let the DeepMind team focus in their original areas of expertise.

- You will have a new (and $$$$) budget to hire computational biology PhDs

- With the new entity, you can fail a lot and not make Verily and Calico look bad (inside and out XYZ)

If you are Verily and/or Calico:

- You don't want a newcomer bring the exciting new projects and ruin your projects, budgets, initiatives

- [very speculative] You don't like Hassabis' progress (or is envy of)

- You think this new project will fail and don't this under your umbrella

precisely this

Does anyone have a good idea what Verily or Calico actually do, or have done?

Verily has done a bunch of things, only a few of them really stuck around. Project Baseline (which pivoted from clinical to covid), Debug, a few spinoffs with other drug companies. Most stuff revolves around physical objects with sensors, intended for a mixture of lab and clinical settings. They make some software, for example Terra (a scientific research platform mostly for genomics/bioinformatics).

Calico does basic life science research and publishes it, https://calicolabs.com/publications and also has private partnerships. They are pretty secretive so it's hard to know but most of the research is about using mouse models to understand fundamental details of aging biology, the long game being to make Larry Page live forever.

Wondering if this could be used to create mass destruction weapons down the line.

Didn't Google just this week say they want to work with the Pentagon again. Coincidence?

> At its most fundamental level, I think biology can be thought of as an information processing system, albeit an extraordinarily complex and dynamic one. Taking this perspective implies there may be a common underlying structure between biology and information science - an isomorphic mapping between the two - hence the name of the company

Reply on news.ycombinator.com