Mol tree tensorize

[NinaCalvi]

Hi,
I am finding a bit confusing these lines from the mol_tree.tensorize function:

for u, v in mol.mol_tree.edges: if len(mol.mol_tree[u][v]['anchor']) > 2: print(mol.smiles) del mol_batch[i-del_num]

why are you deleting this molecule in this instance?
Won't it be used for training anymore?
what exactly does it mean for the mol_tree anchor to be greater than 2?

Thank you!

[ziqi92]

Hi,

This is due to that I found the junction tree algorithm I used cannot decompose some molecules with very complex ring systems as below,

(The molecule above contains a complex ring system with more than two rings and many shared bonds and atoms.)

So I deleted them during the training.

I remembered that in the dataset, there should be only 2 - 5 such molecule pairs. So you may not need to worry about it.

Please let me know if you have any other questions.

Best,
Ziqi

[NinaCalvi]

Thank you!! Quick question about this though - is that you tensorize x and y separately when you load the dataset - so if you delete some of these molecules in the input (x) - would you need to ensure you are deleting the same index in y?
It doesn't seem like you are doing that?

[ziqi92]

Hi,

Thanks for the suggestions. Yes, I think we do need to ensure that these molecules in the molecule pairs are all deleted if it is possible to optimize the molecules by adding or modifying such complex rings. I will fix the issue soon.

Best,
Ziqi