GitLab

We have to integrate the network embeddings produced by @mbertoni using node2vec. These embeddings should be stored as netemb.h5. The format of netemb.h5 files should be the same of sig.h5, i.e. a keys sorted vector, a V matrix (np.float32) (rows sorted accordingly to keys), date, etc.

Present

• As of today, _similarity.py produces similarity vectors stored in /aloy/web_checker/molecules/<IN>/<CH>/INCHIKEY/<similars.h5 or sig.h5>.
• These vectors are currently read by @mbertoni and stored as a network.
• Then, node2vec is ran to produce embeddings.

Future

• The network should be, in principle, loadable from the output of faiss.
• I would recommend that we directly implement this option.

As for the MoA / ATC validations, once the code is integrated within the resource, I will the corresponding functions.