ML task on our project

[theicechol]

We must re-state our ML idea on our data/ molecular fingerprints real soon. Hopefully, we can shape it up by this weekend.

Let's put some information on this issue so that it is easy for us to follow.

• [x] Define the input data : Input a PubChem ID in and get a (list of) KEGG reaction ID out
• [x] Define the extracted information from the input data : Calculate the similarity index (Tanimoto) to the pre-computed common substructure database of enzyme substrates
• [x] Define the query we want to perform to the database : Create our homemade database of promiscuous enzyme with maximal common substructure to be compared with the input compound
• [x] Define the type of data we want from the query of our input and output from the database: We will use chemical similarity for the queries
• [x] Define how to turn those data into a vector/ numerical value for the machine learning methodology: RDKit will take care of that part

[theicechol]

Let me try to point it out First, use the input compound to generate a bunch of compounds based on the similarity index. Then, seek for enzymes that accept one of those substrates for their regular substrate. Finally, go back to the similarity index and rank the enzymatic reaction based on the similarity score.

Still kinda vague, there must be a getter approach!!!

[theicechol]

1) An input data for us should be either SMILES/SMARTS/InChI strings or we can just use a database entry like PubChem entry to call for that particular compound.

No serious problem on managing the input format.

[theicechol]

In KEGG database, the PubChem ID is available which we can direct to that PubChem entry and retrieve the desired string (SMILES or InChI)

[theicechol]

2) From the known enzymes with substrate promiscuity in our database

• take all of their substrate and generate the fingerprint similarity index via RDKit packages --- give numerical value
• Then, get an average of the similarity number of any particular enzyme and generate the substrate promiscuity or substrate tolerance index
• These index will be used further for the search from the input-generated compounds to match with promiscuous enzyme