sandboxnu / cheminformatics Goto Github PK

What kind of fingerprint is rdkit fingerprintMols?

Look into the rdkit cookbook and also research what kinds of fingerprint options there are

how do we do the pains filtering

https://www.rdkit.org/docs/Overview.html#open-source-toolkit-for-cheminformatics

http://chemmine.ucr.edu/

https://github.com/girke-lab/chemminetools

The user has to entire SMILE strings into the website. This issue covers the UI component of the text box and saving the input in some way in the backend.

does it exist? can we scrape it from the catalogue?

Problem: Currently, we have the PAINs filtering implemented, but it is just calling the script from https://github.com/sandboxnu/Lilly-Medchem-Rules. It would be nice to somehow wrap this, like in a python package.

similarity coefficient

https://www.rdkit.org/docs/api-docs.html

particularly looking into
rdkit.Chem.rdShapeHelpers.EncodeShape
rdkit.Chem.rdShapeHelpers.ShapeTanimotoDist
rdkit.Chem.Descriptors
rdkit.Chem.Descriptors3D

Research how each of the measures noted in the academic paper (ROC, PMI) factor into calculation

check if file exists before removing it

how is the page going to look? what is the organization pattern?

Set up ctyoscape.js

Store the parameters in backend so they can reused in future (or for analytics?)

see slack for possible links

(render all mpo nodes as black)

Make sure structure is easy to read and visible

The centroid is the most connected element of the butina cluster algorithm. Every chemical in the cluster is connected to the centroid.

if it has 60% similarity to the centroid

Using rdkit or by other means.

Maybe http://bkchem.zirael.org ->opensource alternative to chemdraw

Clustering takes much longer when you pick a low tanimoto coefficient on alot of data. Warn the user before they make this mistake and assume the program is broken.

Options:
-better error messaging - maybe an error screen with what went wrong
-prevent key error by ignoring duplicate smiles.

We need to create clusters (lists of compounds) to pass to cytoscape to visualize. We could potentially use: https://scikit-learn.org/stable/modules/classes.html#module-sklearn.cluster

framework, django or flask or other TBD

how are we storing it, dictionary ,dataframe ? also, what values are we storing?

Allow users to choose which fingerprint they want to make the app more versatile

If they are going to have like over 60% be singletons

ECFP4 as default

choose a color scheme for the table (think about how the circles were colored in the graph). Maybe have a couple popular/recommended preset options.

(implementation/backend)

without having to go back to the original select options step

hover, labels, etc.

(figure out what we use to cluster)

Similarity coefficient is a number between 0 and 1 describing how closely related inputs have to be in order to be grouped together (degree of similarity). We were thinking of making this an input box with a small scroll wheel or up/down arrows