used RF, XGB, and Logistic Regression w Lasso to identify 20 key genes in AML
-
Data cleaning: merge clinical, manifest, and ensemble gene info for three AML, WT and AML
-
Modeling: modeling on AML data to predict status of "Low Risk" vs. "Not Low Risk" (Standard + High) - if patient shows up more than once, use earlier one
- Lasso Logistic Regression on all 20,000+ features (gene ensembles) --> reduced to 3212
- XGBoost Classifier on all 20,000+ features (gene ensembles) --> 332 "important" features
- RF Classifier on all 20,000+ features (gene ensembles) --> reduced to 1147 "important" features
11 genes deemed important by all three models
Lasso run again until conversion, features reduced to 174 features. 15 shared "important" features between XGBoost and Lasso. Of those 15, 6 belong to original 11, 9 novel.
Those 20 genes warrant further investigation.
| Directory | Description |
|---|---|
| archives | old notebooks |
| data | csvs after merge, in zip file |
| pictures | pictures for readme |
| model_walkthrough.ipynb | notebook for modeling |
| clean.py | cleaning and merging data |
| model_comp.py | tools for modeling, sklearn |
dummies.py not used..