Week ten | Retraining and visualization

GSoC Project

• GSoC Project URL
• Work Repository

Mentor: Augustin Luna (@cannin)

Tasks

1. Re-train DrugCell model.
   Done HP tuning for small data, without the label of Other Score is 0.67.

2. weight visualization I’ve got the correlation score for GO terms and genes. Also add some visualization and explain some drugs.

3. Document Project Dependencies

   Working on modifying README.

4. create datasets for rcellminer

For retraining, I’ve created new file for inputs, cell2ind, drug2ind, and drug2fingerprint.

Comments

Currently working on explaining drugs and correlation from biological phenomena.
Some important GO terms are missing, so it might be challenging to explain whole drugs.
So, I plan to focus on some specific drugs approved by the FDA.
I still need to improve the model, so I will do experiments in parallel.

In addition, I’ve been extracting SMILES data from PubChem ID in rcellminer because I used to use the drug2id from original data.
But right now I’ve created new drug2ind exclusive for our data. So, I can use our own PubChem data directly.
I’ll try this after current experiments.

Next Step

• Explain by biological phenomena.
• Documentation
• add comments
• retraining