| | @Jessica Maat (Deactivated) | JM/SB: General update on wbo work so far and why we narrowed down from a general approach to parameter and chemical series approach due to lack to transferability of parameters to a wider chemistry. ChS: Yeah, I agree that there will be lot of sub-groups and chemistries that are beyond the designed datasets. DM: Yeap, we have spent a good part of last year and early this year to figure out and filter out specific chemistries for interpolated parameters and we realized that there are lot of blocks in terms of general applicability, viz., strong sterics for example, and general chemistry space beyond the designed set.
Some of our current efforts are focused on biaryls, and also checking what the actual distribution of wbos is in the wild. SB: Yeah, we are still assessing whether an interpolated parameter is good or we can just split into two parameters that would outperform a single interpolated one, and such. JM: Here’s The distribution of wbos from ChemBL and QCA with the smarts for biaryls-no-ortho-subs is skewed and located within a short range of wbos from 1 to 1.1.
I looked at several thousands of ChemBL molecules. ChS: It would be great if we can look at the distribution of charges and also use the reasonable tautomer filter from openeye to see if we are missing out any relevant molecules. DM: That’s a good idea, also there would be lot of molecules with salts, that are separated by a period in the smiles, we can cut out that portion and use the molecule smiles. JM: Okay, will check those.
|
|
|
| DM: SB, from these results it looks like we are still seeing distributions of wbos to be in a narrow range, and it’s better if we use JM’s time to continue work on interpolated impropers, do you think the infrastructure is ready? SB: Yeah, that’d be great, may be we can use a similar approach there as well, before jumping into the implementation details. I can do the toolkit implementation for interpolated impropers but we should see the distribution of wbos for those impropers, such as aniline set, among the general set beyond our designed dataset. DM: Yeah, we can do a smarts based search and check on this, does that sound like a path forward JM? JM: Sounds good.
|