Understanding immune-metabolic relationships by modeling the association between Body Mass Index (BMI), adipokines and other immune markers helps to generate novel biological insight. Especially to help target immune-metabolic dysfunctions such as obesity, diabetes or cancer. We received a dataset of 170 immune markers and 902 participants from the Stanford 1000 Immunomes Project. Our aim is to generate a good model to predict BMI. To do so, we have screened different algorithms:
- Linear regression on a reduced feature space
- Regularization algorithms
- PyCaret model selection
- python - version 3.6.5
- R
- pycaret==2.3.10
- markupsafe==2.0.1
- pyyaml==5.4.1 -qq
- scikit-learn==0.23.2
- matplotlib==3.1.3
From feature selection, regularization analysis and the general screening, we can see that model performance is quite similar ( RMSE ∼ 3) and no model offers a very good approximation to BMI yet. Also, in all approaches, leptin, gender and age have come up as the most important features to focus on, as expected and discussed with our mentor.
Project is: finished
CS229 Summer 2022 Paper link