Brainstorming for additional functionalities and features

[JonasSchaub]

Here, we can collect and discuss ideas for new features and functionalities.

Notes from our document so far:

• Functional group frequency analysis based on Ertl alg. - which functional groups appear and how frequently? Grid image of the most frequent ones
• Give CIDER a user interface (web app?)
• More from https://github.com/PatWalters/practical_cheminformatics_tutorials?
• Additional inspiration from Molecule Set Comparator application?

[JonasSchaub]

Make other input file types available?

[OBrink]

I would definitely agree that we should add more input files. And although a web app might be a bit much, it would probably enable access for a lot more people. It does not need to be anything fancy, just some kind of user interface that runs the workflow

[adelenelai]

+1 for SMILES input

[JonasSchaub]

One idea I had: We could add initial structure curation functionalities to CIDER. But I would say only if it can be integrated easily from existing toolkits (like the RDKit ChEMBL curation pipeline, for example). Or we need to add an explicit statement that all curation must be done beforehand.

[JonasSchaub]

@hannbus @Kohulan @OBrink as discussed this morning. Anything to add?

[JonasSchaub]

Idea for a new feature: Generate plots like the ones in Figures 3 and 4 of this paper https://pubs.acs.org/doi/10.1021/acs.jnatprod.8b01022 for the scaffolds and the maybe upcoming functional group analysis (or any kind of fragment analysis).

[OBrink]

But please without this horrible colour scheme! :D

[OBrink]

Parallelisation to speed things up would be useful

[JonasSchaub]

Clustering - e.g. https://rdkit.readthedocs.io/en/latest/Cookbook.html#clustering-molecules could result in a simple bar plot of the clusters and how many molecules are in each cluster, as a quick metric for diversity.

[JonasSchaub]

Wrap dictionary data structure in a class object?

[JonasSchaub]

New idea: Being able to set a "memory mode", i.e. keeping mol objects in memory (e.g. for servers) vs. re-importing the molecules every time (e.g. for laptops).

[JonasSchaub]

Chemical space visualisations with descriptor vectors as input, i.e. every position in the input vector is a specific molecular descriptor value. This is nice because it helps to interpret the principal components in some way based on how much each descriptor value contributes to the respective dimension (loading plot).

[JonasSchaub]

ChemPlot seems to be able to do this using the setting "tailored" instead of "structural".

[JonasSchaub]

https://www.mdpi.com/2218-273X/10/11/1566
See this paper for some more ideas