How to catch activity cliffs? #memo #journal

Activity cliff is very popular concept in medicinal chemistry. It means that similar compounds show very different biological activities. There many approaches to apply the concept to drug design. BTW, what does ‘Similar compounds’ mean? In the area of chemoinformatics 2D fingerprint based similarity is widely used, such as Tanimoto similarity of ECFP, MACCS key

Advertisement

Visualize the torsion drive with different approach #openff #torchani #chemoinformatics #quantum_chemistry

Yesterday, I posted about quantum chemistry based predictive model named ‘torch-ani’. It’s really interesting module which build from lots of QC data. To use torchani API, we can visualize torsion drive with the trained model. It sounds interesting however, I would like to compare the torsion drive results from different approaches. Fortunately, QCArchive provides very

Sample test for quantum ML #pytorch #psikit #RDKit

Recently I have many opportunities to read exciting articles about quantum-machine learning which means some models are trained with quantum chemistry based data such as ANI. I’m interested in the area and fortunately we can use really cool code named torchani. It provides model zoo includes ani-2x ;) Regarding the original article about ANI2x, ANI

Benzenoid substitution pattern of drugs #memo #journal

Phenyl ring is very important parts in medicinal chemistry. So there are many drugs which have benzoid in the substructure. And do you know which is most popular substitution pattern in drugs? Here is an interesting article to answer the question. Here is a link to ACS publication https://pubs.acs.org/doi/abs/10.1021/acs.jmedchem.0c00915 And following link is arxiv, it’s

Compound Generator with Graph Networks, GraphINVENT take2 #chemoinformatics #RDKit #PyTorch

I posted about graph based compound generator named ‘GraphINVENT’ some days ago. https://wordpress.com/block-editor/post/iwatobipen.wordpress.com/3450 Fortunately I could get response from author and could get useful information about their model. The hyperparameter of the model is very important and difficult to optimize but I could get suitable learning rate to train the GDB-13 small set. I changed