Source code for CPAC.network_centrality.utils

from CPAC.pipeline.schema import valid_options
from CPAC.utils.docs import docstring_parameter


[docs]def convert_pvalue_to_r(datafile, p_value, two_tailed=False): ''' Method to calculate correlation threshold from p_value Parameters ---------- datafile : string filepath to dataset to extract number of time pts from p_value : float significance threshold p-value two_tailed : boolean (optional); default=False flag to indicate whether to calculate the two-tailed t-test threshold for the returned correlation value Returns ------- r_value : float correlation threshold value ''' import nibabel as nb import numpy as np import scipy.stats # Get two-tailed distribution if two_tailed: p_value = p_value / 2 # Load in data and number of time pts img = nb.load(datafile).get_data() t_pts = img.shape[-1] # N-2 degrees of freedom with Pearson correlation (two sample means) deg_freedom = t_pts-2 # Inverse Survival Function (Inverse of SF) # Note: survival function (SF) is also known as the complementary # cumulative distribution function (CCDF): F_(x) = p = P(X > x) = 1 - F(x) # The inverse will yield: x = F_^-1(p) = F_^-1(P(X > x)) # where x is a value under the distribution of the random variable X # such that the probability of getting greater than x, is p t_value = scipy.stats.t.isf(p_value, deg_freedom) r_value = np.sqrt(t_value ** 2 / (deg_freedom + t_value ** 2)) # Return correlation coefficient return r_value
# Function to actually do the list merging def merge_lists(deg_list=[], eig_list=[], lfcd_list=[]): merged_list = [] merged_list.extend(deg_list) merged_list.extend(eig_list) merged_list.extend(lfcd_list) # and we're hijacking this function to parse out the files instead! degree_weighted = None degree_binarized = None eigen_weighted = None eigen_binarized = None lfcd_weighted = None lfcd_binarized = None for path in merged_list: if 'degree' in path and 'Weighted' in path: degree_weighted = path elif 'degree' in path and 'Binarize' in path: degree_binarized = path elif 'eigen' in path and 'Weighted' in path: eigen_weighted = path elif 'eigen' in path and 'Binarize' in path: eigen_binarized = path elif 'local_functional' in path and 'Weighted' in path: lfcd_weighted = path elif 'local_functional' in path and 'Binarize' in path: lfcd_binarized = path return (degree_weighted, degree_binarized, eigen_weighted, eigen_binarized, lfcd_weighted, lfcd_binarized) # Separate sub-briks of niftis and save def sep_nifti_subbriks(nifti_file, out_names): ''' ''' import os import nibabel as nib output_niftis = [] nii_img = nib.load(nifti_file) nii_arr = nii_img.get_data() nii_affine = nii_img.get_affine() nii_dims = nii_arr.shape if nii_dims[-1] != len(out_names): if len(nii_dims) == 3 and len(out_names) == 1: pass else: err_msg = 'out_names must have same number of elements as '\ 'nifti sub-briks' raise Exception(err_msg) for brik, out_name in enumerate(out_names): if len(nii_dims) == 3: brik_arr = nii_arr elif len(nii_dims) > 3: brik_arr = nii_arr[:, :, :, 0, brik] out_file = os.path.join(os.getcwd(), out_name+'.nii.gz') out_img = nib.Nifti1Image(brik_arr, nii_affine) out_img.to_filename(out_file) output_niftis.append(out_file) return output_niftis @docstring_parameter(m_options=valid_options['centrality']['method_options'], t_options=valid_options['centrality'][ 'threshold_options']) def check_centrality_params(method_option, threshold_option, threshold): ''' Function to check the centrality parameters. Parameters ---------- method_option : str or int one of {m_options} or index of option threshold_option : str one of {t_options} or index of option threshold: float Returns ------- method_option : str one of {m_options} threshold_option : str one of {t_options} ''' # Check method option if isinstance(method_option, int): if method_option < len(valid_options['centrality']['method_options']): method_option = valid_options[ 'centrality']['method_options'][method_option] else: raise MethodOptionError(method_option) elif not isinstance(method_option, str): raise TypeError('Method option must be a string, but type \'%s\' ' 'provided' % type(method_option).__name__) # Check threshold option if type(threshold_option) is list: threshold_option = threshold_option[0] if type(threshold_option) is int: if threshold_option < len( valid_options['centrality']['threshold_options'] ): threshold_option = valid_options[ 'centrality']['threshold_options'][threshold_option] else: raise ThresholdOptionError(threshold_option, method_option) elif type(threshold_option) is not str: raise TypeError('Threshold option must be a string, but type \'%s\' ' 'provided' % type(threshold_option).__name__) # Format input strings method_option = method_option.lower().rstrip(' ') method_options_v1 = ['degree', 'eigenvector', 'lfcd'] if method_option in method_options_v1: method_option = valid_options['centrality']['method_options'][ method_options_v1.index(method_option) ] if ' ' not in threshold_option: threshold_option = ' '.join([threshold_option, 'threshold']) threshold_option = threshold_option.capitalize().rstrip(' ') # Check for strings properly formatted if method_option not in valid_options['centrality']['method_options']: raise MethodOptionError(method_option) # Check for strings properly formatted if threshold_option not in valid_options['centrality'][ 'threshold_options' ]: raise ThresholdOptionError(threshold_option, method_option) # Check for invalid combinations of method_option + threshold_option if ( method_option == 'local_functional_connectivity_density' and threshold_option == 'Sparsity threshold' ): raise ThresholdOptionError(threshold_option, method_option) # If it's significance/sparsity thresholding, check for (0,1] if ( threshold_option == 'Significance threshold' or threshold_option == 'Sparsity threshold' ): if threshold <= 0 or threshold > 1: raise ThresholdError(threshold_option, threshold) # If it's correlation, check for [-1,1] elif threshold_option == 'Correlation threshold': if threshold < -1 or threshold > 1: raise ThresholdError(threshold_option, threshold) else: raise ThresholdOptionError(threshold_option, method_option) # Return valid method and threshold options return method_option, threshold_option class MethodOptionError(ValueError): """Raised when a selected centrality method option is not supported. """ def __init__(self, method_option): self.method_option = method_option self.message = 'Method option \'%s\' not supported' % method_option super().__init__(self.message) class ThresholdError(ValueError): """Raised when a selected threshold value is not supported for a selected threshold option. """ def __init__(self, threshold_option, threshold): self.threshold_option = threshold_option self.threshold = threshold print(type(threshold)) self.message = f'For \'{threshold_option}\', threshold value must be ' if ( threshold_option == 'Significance threshold' or threshold_option == 'Sparsity threshold' ): self.message += 'a positive number greater than 0 ' elif threshold_option == 'Correlation threshold': self.message += 'greater than or equal to -1 ' else: raise ThresholdOptionError(threshold_option) self.message += 'and less than or equal to 1.\n Currently it is set ' \ f'at {threshold}' super().__init__(self.message) class ThresholdOptionError(ValueError): """Raised when a selected threshold option is not supported for a selected centrality measure. """ def __init__(self, threshold_option, method_option=None): self.method_option = method_option self.threshold_option = threshold_option self.message = f'Threshold option \'{threshold_option}\' not supported' if self.method_option: self.message += ' for network centrality measure ' \ f'\'{method_option}\'' self.message += '; fix this in the pipeline config' if ( method_option == 'local_functional_connectivity_density' and threshold_option == 'Sparsity threshold' ): valid_options = ' or '.join([ f"'{t}'" for t in valid_options[ 'centrality' ]['threshold_options'] if t != threshold_option ]) self.message += f'. \'{method_option}\' must use {valid_options}.' super().__init__(self.message)