2024-01-09

This is Marco's daily open-notebook.

Today is 2024.01.09

Interesting video for managing big projects in python : https://www.youtube.com/watch?v=tAuRQs_d9F8&t=169s

We currently have the LOTUS database and this must not chnage by adding new molecules by downloading CHEBI or NCBI or whatever.

Then maybe be able to add new species and new molecules (see question below with inductive)

Why can't I merge a graph with edge weights with an other that has no weights. Why can't we do that ever if the edges are of different type ?
Since we have a graph that evolves quickly, what are the algorithm that are inductive and not transductive. I personally used GraphSAGE but it is supposed to work only on homogenous graphs which is not our case.
Can we think about making predictions of higher order such as linking molecular pathways WITH Genus or families of organisms ?
How can I train the model ?
Should I train/test split on heterogenous nodes only ?
Should I run the edge prediction pipeline ?
What kind of embedding should we go for ?
How can I perform grid search to look for the best embedding on the cluster ? I was able to run stuff on the cluster but when I put them as list it seems like
How can I train the model to only train and test edges that are biolink:in_taxon ?
How can I use edges weights for the edge predictions ?