Protein shapes in 3D…tbh, doesn’t really look like chicken (Source: DeepMind)
Proteins are much more than the chicken in the kale salad you’re eating right now.
At the most basic level, they are molecules that influence how bacteria, viruses, and the human body behave.
And the key way of understanding how a protein works is by knowing its 3D shape, which dictates how it interacts with other proteins.
Last week, DeepMind — an AI startup acquired by Google for $500m in 2014 — released the predicted shapes of 350k+ proteins, per The New York Times.
This freely available knowledge drop is powered by a DeepMind AI program called AlphaFold that can analyze huge dumps of data and draw insights.
In the olden times (pre-2014), the process of identifying a protein’s shape was brutally long: It could take years and required X-rays and microscopes.
Per NYT, the predicted protein shapes has many applications:
… so the protein prediction still needs to be paired with physical experiments.
Still, one professor tells NYT that the broad protein understanding — including ~250k new shapes — is 10 years ahead of schedule.
Google has no plans to commercialize the product. Not that it really needs to: The search giant made $147B on ads last year… including a ton for chicken kale Caesar salads, we’re sure.