Awesome

Genetically determined susceptibility to malaria

Report describing the work is in report.pdf file.

To reproduce our results:

1. Download dataset from here and put it under "data" directory in the repo root.
2. Create conda environment from environment.yml and activate it by running:

   conda env create -f environment.yml
   conda activate ntds_2019 
   

3. Execute the jupyter notebooks in the following order:
   1. baseline_phenotype_predictor.ipynb
   2. coexpression_graph_construction.ipynb
   3. coexpression_graph_exploration.ipynb
   4. SNP_expression_imputation.ipynb

Stages of the project

1. Exploratory Data Analysis
   • Obtain PhenoID for diseases of interest (Malaria and Influenza)
   • Get phenotype value associated to the PhenoID of interest for each mouse
   • Check distribution of phenotype values obtained previously
   • Select single phenotype for simplicity: malaria susceptibility
   • Normalize the data: subtract mean and divide by standard deviation
   • Building a dataframe of all (SNP expression, tissue) pairs for all mice for constructing the coexpression graph
2. Setting a baseline
   • Building a regression model, ridge regression and random forests, to predict phenotype values given genes expression. We take all expressions and filling missing values with mean
3. Building co-expression graph
   • (Genes, Tissue) as node and we use the expression values of each mouse to compute the distance metric.
4. Imputation of SNP expression on co-expression graph using Tikhonov filter
   • Smoothing for filling missing values for SNP expression
5. Prediction of phenotype values using additional expression data from 4.
   • Using the best method, comparing with baseline.