Uchicago Summer Session 2023 Datascience Final Project

By Alex Ramirez, James Lin, Corona Tian and Jiaxi Gao

Predicting solubility of molecules using different models and featurizers

• Google slides here
• Writeup here (WIP)

Building (poetry)

1. Download Poetry
2. Clone this repository
3. Run poetry install in the root directory of this repository
4. Either
   1. Option 1, jupyter lab
      1. Run poetry run jupyter notebook to start the notebook server
      2. Open the outputted link in your browser
   2. Option 2, vscode
      1. Run poetry env info --path
      2. Open the command pallet (Ctrl+Shift+P) and type Python: Select Interpreter
      3. Click "Enter interpreter path" and paste the path from step 1
      4. Open the jupyter notebook file in vscode
      5. On the top right, click the "Select Kernel" button and select the kernel with the same name as the path from step 1

Building (pip) (untested)

1. Make sure you running a python ver >=3.10,<3.11
2. Clone this repository
3. Run pip install -r requirements.txt in the root directory of this repository
4. Follow step 4 from the poetry instructions

Notebooks (in /notebooks)

Notebook	Description
main.ipynb	Main notebook. Data exploration, analysis, featurizer & model training / testing
outliers.ipynb	More data exploration, analysis & seeing outliers in the dataset
testing_models.ipynb	Testing saved models from main.ipynb on new, never seen molecules
util.py	Utility functions for saving / loading saved models

Models (in /models) (load using pickle.load)

File	Name	Time	Mean 5-fold RMSE	Mean 5-fold r^2	RMSE	r^2
random_forest.pkl	Random forest	23.2s	1.1444	0.7599	1.1474	0.7683
bagging.pkl	Bagging (Random forest estimator)	1m 14.2s	1.1461	0.760	1.1529	0.7660
bagging_reg.pkl	Bagging (Decision tree estimator)	21.6s	1.1466	0.7563	1.1565	0.7646
gradient_boosting.pkl	Gradient boosting	12.7s	1.1767	0.7443	1.2129	0.7411
decision_tree.pkl	Decision tree	3.1s	1.3269	0.6747	1.3840	0.6629
adaboost.pkl	AdaBoost	7.2s	1.4032	0.6358	1.4327	0.6387

Datasets (in /data)

File	Source	Description
aqsoldb.csv	https://www.kaggle.com/datasets/sorkun/aqsoldb-a-curated-aqueous-solubility-dataset	Main training dataset
delaney.csv	https://pubs.acs.org/doi/10.1021/ci034243x	18 extra unique molecules for testing models