Maximum Flow Formulation Identifies High-Confidence Variants in Simple Repeat Sequences Associated with Autism Spectrum Disorder
This repository contains the source code for implementing maximum flow for dimensionality reduction and improved feature stability in high-dimensional whole genome datasets. The algorithm uses the presence of linkage disequilibrium (LD) to cluster similar variants and reduce the size of the feature space.
This code base is implemented in Python 3.6.1 and requires the following libraries: numpy (1.16.4). Other necessary libraries include: bcftools (1.8), plink2 (2.0).
Begin by performing k-fold cross-validation with any l1-regularized machine learning algorithm on a high-dimensional whole genome dataset. In this implementation, we use k=5. Isolate all predictive variants (nonzero coefficient scores) from each fold and create an input folder with k files, each in the following tab-delimited format:
chr-variant1 coefficient_score
chr-variant2 coefficient_score
...
First, we will generate a formatted list of all variants from the cross fold analysis. Run the following script:
python step_0.py <input_folder>
This will generate two files in the input folder: variant_ranges.tsv and vcf_input.txt. Next, run the following command to filter these variants from whole genome vcf files:
bcftools view ${vcf_filename} -o ${outputfilename} --output-type z --regions-file vcf_input.txt
Now, we use plink to generate the necessary files for computing LD. This should create a series of files, include .bim and .bed files.
plink2 --vcf $vcf_filename --extract range variant_ranges.tsv --make-bed --out $prefix
Run the following script to assign id numbers to variants so that LD scores can be computed. Specify the prefix of the .bim and .bed files (defined in the previous command).
python step_1.py <prefix>
Next, we use this information to calculate pairwise LD scores for variants across different folds. $num1 and $num2 represent the fold numbers to be compared (e.g. 1 and 3).
srun ipython step_2.py $input_folder $prefix $num1 $num2
Finally, run the maximum flow algorithm to identify clustered groups of variants in LD.
python maxflow.py 1_2_3_4_5
Results are stored in "results.txt" and "path.txt".
Validation code to set up the machine learning classifier is provided in validation.ipynb. Please note that some cells are nonfunctional, since large datasets cannot be uploaded to Github. Thus, please modify the code and dependent filepaths to suit your needs.
- Maximum flow set-up and Ford-Fulkerson code adapted from: https://www.geeksforgeeks.org/ford-fulkerson-algorithm-for-maximum-flow-problem/