/// <summary>
/// Imputes the missing clinical data values for samples.
/// </summary>
// imputation follows the concept below:
// we use the average value of all clinical data for a given phenotype
// in case there is only one instance of the clinical data for the phenotype and cannot be imputed the we just assign 1.
private static void impute_clinicalDataForSamples(List <clinicalDataFS> listTmp_clinicalDataFS)
{
clinicalDataFS cdf;
Dictionary <string, imputedValues> numcov;
List <clinicalDataFS.sampleWeight> swcov;
foreach (clinicalDataFS cdfs in listTmp_clinicalDataFS)
{
//Impute numerical covariates
numcov = new Dictionary <string, imputedValues>();
foreach (KeyValuePair <string, imputedValues> kvp_s_iv in cdfs.Numerical_covariates)
{
if (kvp_s_iv.Value.Non_imputed == -1)
{
if (listTmp_clinicalDataFS.Where(x => x.Phenotype == cdfs.Phenotype).SelectMany(x => x.Numerical_covariates)
.Where(x => x.Key == kvp_s_iv.Key).Select(x => x.Value.Non_imputed).Count(x => x != -1) == 0)
{
outputToLog.WriteWarningLine(kvp_s_iv.Key + " for phenotype " + cdfs.Phenotype + " could not be imputed! Too few records!");
numcov.Add(kvp_s_iv.Key, new imputedValues()
{
Imputed = 1,
Non_imputed = -1
});
}
else
{
numcov.Add(kvp_s_iv.Key, new imputedValues()
{
Imputed = listTmp_clinicalDataFS.Where(x => x.Phenotype == cdfs.Phenotype).SelectMany(x => x.Numerical_covariates)
.Where(x => x.Key == kvp_s_iv.Key).Select(x => x.Value.Non_imputed).Where(x => x != -1).Average(),
Non_imputed = -1
});
}
}
else
{
numcov.Add(kvp_s_iv.Key, new imputedValues()
{
Imputed = kvp_s_iv.Value.Imputed,
Non_imputed = kvp_s_iv.Value.Non_imputed
});
}
}
//Impute sample weights
swcov = new List <clinicalDataFS.sampleWeight>();
foreach (clinicalDataFS.sampleWeight kvp_sw_iv in cdfs.SampleWeight_covariates)
{
if (kvp_sw_iv.weight.Non_imputed == -1)
{
if (listTmp_clinicalDataFS.Where(x => x.Phenotype == cdfs.Phenotype).SelectMany(x => x.SampleWeight_covariates)
.Where(x => x.tissue == kvp_sw_iv.tissue && x.charge == kvp_sw_iv.charge)
.Select(x => x.weight.Non_imputed).Count(x => x != -1) == 0)
{
outputToLog.WriteWarningLine(kvp_sw_iv.tissue + " " + kvp_sw_iv.charge + " for phenotype " + cdfs.Phenotype + " could not be imputed! Too few records!");
swcov.Add(new clinicalDataFS.sampleWeight()
{
tissue = kvp_sw_iv.tissue,
charge = kvp_sw_iv.charge,
weight = new imputedValues()
{
Imputed = -1,
Non_imputed = -1
}
});
}
else
{
swcov.Add(new clinicalDataFS.sampleWeight()
{
tissue = kvp_sw_iv.tissue,
charge = kvp_sw_iv.charge,
weight = new imputedValues()
{
Imputed = listTmp_clinicalDataFS.Where(x => x.Phenotype == cdfs.Phenotype).SelectMany(x => x.SampleWeight_covariates)
.Where(x => x.tissue == kvp_sw_iv.tissue && x.charge == kvp_sw_iv.charge)
.Select(x => x.weight.Non_imputed).Where(x => x != -1).Average(),
Non_imputed = -1
}
});
}
}
else
{
swcov.Add(new clinicalDataFS.sampleWeight()
{
tissue = kvp_sw_iv.tissue,
charge = kvp_sw_iv.charge,
weight = new imputedValues()
{
Imputed = kvp_sw_iv.weight.Imputed,
Non_imputed = kvp_sw_iv.weight.Non_imputed
}
});
}
}
cdf = new clinicalDataFS();
cdf.setClinicalDataFromVariable(cdfs.Id, cdfs.Phenotype, swcov, cdfs.Categorical_covariates, numcov, cdfs.Ignored_covariates);
List_clinicalData.Add(cdf);
}
}
/// <summary>
/// Reads the file that contains the clinical data for the samples.
/// </summary>
public static void Read_ClinicalDataForSamples()
{
List <clinicalDataFS> listTmp_clinicalDataFS = new List <clinicalDataFS>();
//set numeric and categorical metadata dictionary indecies
//set id index
//set ignored covariates
//set decision
int id_index = -1, decision_index = -1;
Dictionary <string, int> numerical_covariates = new Dictionary <string, int>(), categorical_covariates = new Dictionary <string, int>(),
ignored_covariates = new Dictionary <string, int>(), sampleWeight_covariates = new Dictionary <string, int>();
using (TextReader input = new StreamReader(@"" + publicVariables.clinicalDataFile))
{
#region set covariate indecies
List <string> covariate_names = input.ReadLine().Split(publicVariables.breakCharInFile).Select(x => x.ToLower().Replace("-", "_").Replace(" ", "_").Replace("=", "_")).ToList(),
covariate_types = input.ReadLine().Split(publicVariables.breakCharInFile).Select(x => x.ToLower()).ToList();
#region check headers and types of metadata
if (covariate_names.Count != covariate_types.Count)
{
outputToLog.WriteErrorLine("Meatdata file contains unequal number of covariates and types");
}
if (covariate_types.Count(x => x != "numeric" && x != "categorical") > 0)
{
outputToLog.WriteErrorLine("Meatdata file contains unexpected types of entries. They should be strictly numeric or categorical");
}
if (covariate_names.Count != covariate_names.Distinct().Count())
{
outputToLog.WriteErrorLine("Meatdata file contains duplicates in the header. There should be only unique entries");
}
#endregion
foreach (string covar_name in covariate_names)
{
if (publicVariables.sampleId == covar_name)
{
if (id_index == -1)
{
id_index = covariate_names.IndexOf(covar_name);
}
else
{
outputToLog.WriteWarningLine("CovarId index had been already set once. The second entry will be ignored. The metadata/covariates file contains duplicates");
}
}
else if (publicVariables.phenotype == covar_name)
{
if (decision_index == -1)
{
decision_index = covariate_names.IndexOf(covar_name);
}
else
{
outputToLog.WriteWarningLine("CovarDecision index had been already set once. The second entry will be ignored. The metadata/covariates file contains duplicates");
}
}
else if (publicVariables.normCovars.Any(x => x == covar_name))
{
if (covariate_types.ElementAt(covariate_names.IndexOf(covar_name)) == "numeric")
{
numerical_covariates.Add(covar_name, covariate_names.IndexOf(covar_name));
}
else if (covariate_types.ElementAt(covariate_names.IndexOf(covar_name)) == "categorical")
{
categorical_covariates.Add(covar_name, covariate_names.IndexOf(covar_name));
}
else
{
outputToLog.WriteErrorLine("Meatdata file contains unexpected types of entries. They should be strictly numeric or categorical");
}
}
else if (publicVariables.sampleWeight.Any(x => x.Item1 == covar_name))
{
if (covariate_types.ElementAt(covariate_names.IndexOf(covar_name)) != "numeric")
{
outputToLog.WriteErrorLine("Sample weight columns in the meatdata file should be strictly numeric");
}
sampleWeight_covariates.Add(publicVariables.sampleWeight.Where(x => x.Item1 == covar_name).Select(x => x.Item2 + "_" + x.Item3).First(),
covariate_names.IndexOf(covar_name));
}
else
{
ignored_covariates.Add(covar_name, covariate_names.IndexOf(covar_name));
outputToLog.WriteWarningLine(covar_name + " was not mentioned in the metadata/covariates file. Assuming ignore");
}
}
#region comment out
//for (int i = 0; i < covariate_names.Count; i++)
//{
// if (covariate_types[i].ToLower() == "ignore")
// {
// outputToLog.WriteWarningLine("ignored covariate: " + covariate_names[i].ToLower());
// if (ignored_covariates.ContainsKey(covariate_names[i]))
// {
// outputToLog.WriteErrorLine("Covariate " + covariate_names[i] + " already exists");
// }
// else
// {
// ignored_covariates.Add(covariate_names[i].ToLower(), i);
// }
// }
// else if (covariate_types[i].ToLower() == "id")
// {
// if (id_index == -1)
// {
// id_index = i;
// }
// else
// {
// outputToLog.WriteWarningLine("ID index had been already set once! Cannot reset it!");
// }
// }
// else if (covariate_types[i].ToLower() == "decision")
// {
// if (decision_index == -1)
// {
// decision_index = i;
// }
// else
// {
// outputToLog.WriteWarningLine("Decision index had been already set once! Cannot reset it!");
// }
// }
// else if (covariate_types[i].ToLower() == "numeric")
// {
// if (numerical_covariates.ContainsKey(covariate_names[i].ToLower()))
// {
// outputToLog.WriteErrorLine("Covariate " + covariate_names[i] + " already exists");
// }
// else
// {
// numerical_covariates.Add(covariate_names[i].ToLower(), i);
// }
// }
// else if (covariate_types[i].ToLower() == "categorical")
// {
// if (categorical_covariates.ContainsKey(covariate_names[i].ToLower()))
// {
// outputToLog.WriteErrorLine("Covariate " + covariate_names[i] + " already exists");
// }
// else
// {
// categorical_covariates.Add(covariate_names[i].ToLower(), i);
// }
// }
// else if (covariate_types[i].ToLower() == "sampleweight")
// {
// if (sampleWeight_covariates.ContainsKey(covariate_names[i].ToLower()))
// {
// outputToLog.WriteErrorLine("Covariate " + covariate_names[i] + " already exists");
// }
// else
// {
// sampleWeight_covariates.Add(covariate_names[i].ToLower(), i);
// }
// }
// else
// {
// outputToLog.WriteErrorLine("Unknown type " + covariate_types[i] + " covariate");
// }
//}
#endregion
#endregion
string line;
clinicalDataFS cdfs;
while ((line = input.ReadLine()) != null)
{
cdfs = new clinicalDataFS();
cdfs.setClinicalDataFromFile(id_index, decision_index, sampleWeight_covariates, categorical_covariates, numerical_covariates, ignored_covariates, line);
listTmp_clinicalDataFS.Add(cdfs);
}
}
impute_clinicalDataForSamples(listTmp_clinicalDataFS);
}