Presenting his work chemper and graphs to support R groups
Numbers on nodes are unique identifiers. These are minimal “atomic” identifiers. Edges in the graph have bond orders
alingnment of graphs. In this case requires “alignment of smarts patterns”
Graph vs SMARTS hierarchy. What is the hierarchy of some patterns. Which are subsets of each other?
CBy: E and F should be switched?
Slide: Comparing non-bijecxtive graphs
CBy: Could we use something else to describe connectity of the graph?
TG: There is lots of implicit connectivity information in the patterns
CBy: Maybe we don’t have to worry about what the degree is.
CBy: Really like your approach overall. Looks pretty good. I’m thinking “how to break it”. Thinking about the alketoh sandbox, has ethers and alcohols. Can you apply this usefully starting with bunch of alkethoh molecules to seperate out bonds, angles, etc. Worried about all possible permutations, curse of dimensionality. Maybe introduce bias? Use external ( ?) information to include here as sort of bias?
TG: Curse of dimensionality. Graph appraoach makes things a bit slower. Mapping must be computed only once per parameter. It is generally still quick. FB is still the bottleneck it seems. Still must apply it to large dataset.
CBy: Is your current dataset still just alkanes?
TG: yes.
CBy: Going forward we need different datasets for different aspects for things like bonds, dependencies of things “two atoms away” in cases of amide/urea/etc… Kind of as in differenty assays in biochem.
TG: Yes, we can try different things like that.
TG: Can we avoid recursive patterns? Don’t how to manipulate them?
CBy: How far can we get without them?
TG: Should I take them out the force field?
CBn: Take them and try to learn them with Sirkky/Chemper (? ) and see if you can get patterns without recursive patterns?
CBy: My initial idea is to see how far we can go without recursive patterns. Try to leave them out and see if they become necessary.
CBn: One could have specific recursive patterns as decorators and try to use them as substitute for the recursive patterns. We tried that some time ago.
