Protein Data Bank

I feel like most people don't actually enjoy doing a lot of the grant finding and reading into the details and finding the specifics, so if you have a block of sub agents that can really specify their actions and their results to only grant finding, for instance, right, then scientists can go and do the experimentation work that they actually care about.

One of the problems in crystallography that I learned a lot a long time ago was these unstructured domains, right? So PDBs bu- based on these structured crystallized da- uh, proteins.

But there's these, there are these things called IDPs or intrinsically disordered proteins.

So these are domains that we don't really know how they behave in cells, right? So w- when you go to AlphaFold, you can see that the prediction scores are very s- very low for those.

I'd like to be able to build a viewer for the protein structure that we crystallized.

The PDB file lives inside the data directory.

It's got some waters inside there represented, so I'd like to not visualize the waters, but the other head atoms which are which correspond to the substrate that we co-crystallized with the protein, I would like to visualize the substrate as balls and sticks, and I would like to visualize the protein, both chains, A and B, as a ribbon.

So help me make an AnyWidget viewer that uses 3DMold.js to visualize that particular protein and we'll do one more thing with that structure after you make it work.

And the data, the AI is in the end only as good as the data that trains it.

And we're generating... special data in the structural protomic space, while most of our competition is just retraining the same bank of data in the PDB in different qualities, right? If you look at AlphaFold Bolts, Becker Lab, and so on and so forth, they're all competing on the same infrastructure of the crystal structures.

But before I elaborate about the special data we're generating, maybe I can give you a few words about this problem we're trying to solve.

So we're an AI-induced proximity company.

And this is what we understood is this is an area where the traditional co-folding models usually fail.

They're either not trained by the right data and they're overconfident.

And since unlike single proteins, there's not enough structural data in the PDB, to become a training set for these protein complexes.

And if you go to two or more proteins, they also have endless conformations that they can fit to.