Repo for loading and analyzing the data from the Allen Institute of Brain Science's Aging, Dementia, & TBI Study. The goal of the project is to develop features from gene expression levels for over 50,000 genes to be used along with neuropathological measurements, demographic information, and medical history data from 107 donors in models of dementia probability. Most scripts were written using Anaconda distributions of R 3.5.1 and Jupyter Lab. There are a few Python versions that I've played with but the bulk of the project is in R.
Inputs and outputs of scripts in ETL
, EDA
, feature-engineering
, final-dataset-prep
, and lasso-models-and-evaluation
are stored in data
. The work
folder contains scripts in progress.
The blog-notebooks
folder has some Jupyter notebooks I made in the process of blogging about this project. See my Jekyll blog.
This project is being used to fulfill the requirements of the Master of Science in Predictive Analytics program (now Master's in Data Science) from Northwestern University. Thanks for checking it out!
expected_count_TPM_FPKM_load.py/.R
- Loads data from study website (R and Python versions)subset_count_data.R
- Subsets RNA-seq counts by brain region or donor sex
plotly_mds_plots_html.R
- Generates HTMLwidget multidimensional scaling plots of RNA-seq data using theplotly
librarydifferential_expression_dispersions_exact_tests.R
- hypothesis testing for differential gene expressiondifferential_expression_visualizations.R
- visualizations & analysis of differential analysis results
clustering_experiments_gene_expression.R
- UsesclValid
to test different clustering methods and parameters on gene expression levels datafeature_engineering_dataset_construction
- Runs k-medoids clustering via CLARA; adds medoids as features to tables of neuropathological, demographic, and medical history data for each patient/donor; accesses the mygene.info database for gene ontologies and creates cloud visualizations of biological process terms for each cluster
data_preparation.R
- Preparation of final dataset for modeling, including: dropping variables, converting categorical features to factor variables & reordering levels, and imputation of missing variables by CART (classification & regression trees) using themice
library; also includes location metric (mean & median) comparisons for all numeric variables