Dear all, I'm hoping for a little bit of guidance.
I'm using RDKit and sklearn to implement a random forest model trained using our drug screening (1600 drugs) results to predict new drug responses.
I'm familiar with the Python RDKit code to generate different fingerprints and the sklearn code to implement the model.
However, I would appreciate any links to published works or technical papers that clearly demonstrate how to divide the dataset (for this particular problem i.e., some_Y_response ~ X_fingerprint) up into training, testing and validation sets.
I've gone through a lot of publications, typically coming from journals such as the Journal of Cheminformatics, and I'm quite shocked by the amount of missing information concerning the training of regression/classifier models.
Many thanks
Anthony
Dear Anthony Nash
These articles might be an asset, have a look:
1. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0219774
2. https://www.nature.com/articles/s41698-020-0122-1
3. https://pubs.acs.org/doi/10.1021/acs.jcim.9b00236
4. https://www.sciencedirect.com/science/article/pii/S088875431830466X
5. https://jcheminf.biomedcentral.com/articles/10.1186/s13321-018-0304-9
6. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6622537/
Kind Regards
Qamar Ul Islam
In its most basic form, it's as simple as making a (let's say) 80:10:10% random split of your dataset for training, testing and validation.
If your molecules can be uniquely divided into some specific classes or categories, you might have to look into "stratified sampling" to draw your samples.
Of course, If you work with a single 80:10:10% TTV split, then you lose 20% of your precious data that could be used to train the model. You can salvage this by doing "k-fold cross validation", where you divide your data set into k blocks of equal size and then retrain your model multiple times, each time using different blocks for training and testing.
This post should act as a good entry point:
https://www.mygreatlearning.com/blog/cross-validation
When it comes to sklearn, the model_selection module has the functions you need:
https://scikit-learn.org/stable/modules/classes.html#module-sklearn.model_selection
- Badges
- Science topic
- Similar topics
- Biological Science
- Molecular Biology
- Biomolecules
- Nucleic Acids
- RNA