Permutation Tests for Cross CNV Paper

This repository contains all the scripts and notebooks for generating null models of betamaps through permutation and testing for significance.

Method summary:

• For each contrast (continuous scores and case-control) generated a null model of 5000 pseudo-betamaps
  • Case-control: Random permutation of case-control labels and generate betamap
  • Continuous scores: Random permutation of score values and generate betamap
• Significance of mean shift and variance
  • For each contrast, compared mean shift (variance) of actual betamap to mean shift (variance) of the 5000 pseudo-betamaps to get pvalue
• Significance of mirror effect (opposite mean shifts)
  • For loci with both DEL & DUP in dataset, for each pair (contrast1, contrast2):
    • Generate null model of differences b/w mean shifts
    • Compare difference b/w mean shift of actual betamaps for contrast1 and contrast2 with distribution of differences to get pvalue
• Significance of correlations
  • For each pair (contrast1, contrast2):
    • Generate null model of correlations
      • Compute correlation between 5000 pairs of betamaps from null models of contrast1 and contrast2
    • Compare difference b/w correlation of actual betamaps for contrast1 and contrast2 with distribution of correlations to get pvalue

Scripts:

• generate_betamaps.py
  • Generate the betamaps for each case-control pair and variable effect.
  • Arguments:
    • path_pheno
      • Path to the phenotype .csv file w/ cases one hot encoded.
    • path_connectomes
      • Path to connectomes .csv file w/ connectomes in upper triangular form.
    • path_out
      • Path to an output directory.
• generate_null_model.py
  • Generate 5000 pseudo-betamaps to form a null distribution for a case-control pair.
  • Arguments:
    • case
      • Which case to run the null model for, must be in:
      • ['IBD', 'DEL1q21_1', 'DEL2q13', 'DEL13q12_12', 'DEL15q11_2', 'DEL16p11_2', 'DEL17p12', 'DEL22q11_2', 'TAR_dup', 'DUP1q21_1', 'DUP2q13', 'DUP13q12_12', 'DUP15q11_2', 'DUP15q13_3_CHRNA7', 'DUP16p11_2', 'DUP16p13_11', 'DUP22q11_2', 'SZ', 'BIP', 'ASD', 'ADHD'].
    • path_pheno
      • Path to the phenotype .csv file w/ cases one hot encoded.
    • path_connectomes
      • Path to connectomes .csv file w/ connectomes in upper triangular form.
    • path_out
      • Path to an output directory.
• generate_null_model_continuous.py
  • Generate 5000 pseudo-betamaps to form a null distribution for a variable effect.
  • Arguments:
    • contrast
      • Which contrast to run.
    • path_pheno
      • Path to the phenotype .csv file w/ cases one hot encoded.
    • path_connectomes
      • Path to connectomes .csv file w/ connectomes in upper triangular form.
    • path_out
      • Path to an output directory.
• significance_corr.py
  • Get the significance for correlation b/w betamaps.
  • Arguments:
    • n_path_mc
      • Path to directory w/ mean corrected null models (output dir of generate_null_model.py and generate_null_model_continuous.py).
    • b_path_mc
      • Path to directory w/ mean corrected betamaps (output dir of generate_betamaps.py).
    • path_out
      • Path to an output directory.
    • path_corr
      • Path to directory w/ correlations & correlation distributions (intermediate outputs). If re-running script just for matrices, saves the heavy step of computing the distributions, should be the same as path_out.
  • Notes:
    • Needs to be complete to run i.e. check that all betamaps and null models exist in given directories for cases/contrasts in ilnes 125-130.
    • Edit the subsets according to what cases/contrasts are desired in each figure. FDR values are determined based on the cases/contrasts in the subset.
• significance_mean_shift_var.py
• significance_mirror_effect.py