@Override
public void run() {
boolean poison = false;
while (!poison) {
try {
WorkPackage pack = m_workpackage_queue.take();
if (!pack.getPoison()) {
analyze(pack);
} else {
poison = pack.getPoison();
}
} catch (InterruptedException ex) {
ex.printStackTrace();
}
}
// if (rConnection != null) {
// rConnection.close();
// }
}
private static void determineFoldchange(
double[] genotypes, double[] expression, Result r, int d, int p, WorkPackage wp) {
int numAA = 0;
int numBB = 0;
double sumAA = 0;
double sumBB = 0;
for (int i = 0; i < genotypes.length; i++) {
// if (indWGA[i] != -1) {
if (genotypes[i] == 0) {
sumAA += (expression[i] * 2);
numAA += 2;
}
if (genotypes[i] == 1) {
sumAA += (expression[i]);
sumBB += (expression[i]);
numAA++;
numBB++;
}
if (genotypes[i] == 2) {
sumBB += (expression[i] * 2);
numBB += 2;
}
// }
}
sumAA /= (double) numAA;
sumBB /= (double) numBB;
double min = sumAA;
if (sumBB < min) {
min = sumBB;
}
// normalize to mean of 1
if (min < 0) {
sumAA += Math.abs(min) + 1;
sumBB += Math.abs(min) + 1;
}
if (wp.getFlipSNPAlleles()[d]) {
r.fc[d][p] = sumAA / sumBB;
} else {
r.fc[d][p] = sumBB / sumAA;
}
}
private void convertResultsToPValues(WorkPackage wp, Result dsResults) {
// per probe, convert to p-value
int numProbes = dsResults.zscores[0].length;
boolean hasResults = false;
for (int p = 0; p < numProbes; p++) {
int nrDatasetsPassingQC = 0;
int nrTotalSamples = 0;
double zSum = 0;
double betasum = 0;
for (int d = 0; d < m_numDatasets; d++) {
// TODO: check whether this actually returns the correct Z-scores: should the stored values
// be flipped?
double zscore = dsResults.zscores[d][p];
double correlation = dsResults.correlations[d][p];
Integer numSamples = dsResults.numSamples[d];
if (!Double.isNaN(correlation)) {
boolean flipalleles = wp.getFlipSNPAlleles()[d];
if (m_useAbsoluteZScores) {
zscore = Math.abs(zscore);
dsResults.zscores[d][p] = Math.abs(zscore);
dsResults.correlations[d][p] = Math.abs(correlation);
if (!Double.isNaN(dsResults.beta[d][p])) {
dsResults.beta[d][p] = Math.abs(dsResults.beta[d][p]);
}
wp.getFlipSNPAlleles()[d] = false;
} else if (flipalleles) {
zscore = -zscore;
// dsResults.zscores[d][p] = zscore;
// correlation = -correlation;
// dsResults.correlations[d][p] = correlation;
}
nrDatasetsPassingQC++;
double weight = Descriptives.getSqrt(numSamples);
zSum += (zscore * weight);
nrTotalSamples += numSamples;
if (determinebeta) {
if (flipalleles) {
betasum += (-dsResults.beta[d][p] * numSamples);
dsResults.beta[d][p] = -dsResults.beta[d][p];
} else {
betasum += (dsResults.beta[d][p] * numSamples);
}
}
}
}
if (nrTotalSamples > 0
&& (testSNPsPresentInBothDatasets && nrDatasetsPassingQC == m_numDatasets)
|| (!testSNPsPresentInBothDatasets && nrDatasetsPassingQC > 0)) {
testsPerformed++;
hasResults = true;
double sqrtSample = Descriptives.getSqrt(nrTotalSamples);
double zScore = zSum / sqrtSample;
double pValueOverall = Descriptives.convertZscoreToPvalue(zScore);
dsResults.pvalues[p] = pValueOverall;
dsResults.finalZScore[p] = zScore;
// determine assessed allele....
if (determinebeta) {
betasum /= nrTotalSamples;
double metase = 1 / Math.sqrt(nrTotalSamples);
dsResults.finalBeta[p] = betasum;
dsResults.finalBetaSe[p] = metase;
}
} else {
dsResults.pvalues[p] = Double.NaN;
dsResults.finalZScore[p] = Double.NaN;
}
}
wp.setHasResults(hasResults);
wp.setResult(dsResults);
}
private void analyze(WorkPackage wp) {
testsPerformed = 0;
currentWP = wp;
wp.setNumTested(0);
// RunTimer t1 = new RunTimer();
// load SNP genotypes
SNP[] snps = wp.getSnps();
int[] probes = wp.getProbes();
Result dsResults = null;
double[] snpvariances = new double[m_numDatasets];
double[][] snpmeancorrectedgenotypes = new double[m_numDatasets][0];
double[][] originalgenotypes = new double[m_numDatasets][0];
boolean[][] includeExpressionSample = new boolean[m_numDatasets][0];
for (int d = 0; d < m_numDatasets; d++) {
SNP dSNP = snps[d];
if (dSNP != null) {
double[] x = dSNP.selectGenotypes(m_expressionToGenotypeIds[d], false, true);
originalgenotypes[d] = dSNP.selectGenotypes(m_expressionToGenotypeIds[d], false, false);
int xLen = x.length;
double meanX = JSci.maths.ArrayMath.mean(x);
snpmeancorrectedgenotypes[d] = new double[xLen];
for (int i = 0; i < xLen; i++) {
snpmeancorrectedgenotypes[d][i] = x[i] - meanX;
}
double varianceX = JSci.maths.ArrayMath.variance(x);
if (varianceX != 0) {
snpvariances[d] = varianceX;
int inds[] = m_expressionToGenotypeIds[d];
int sampleCount = m_expressionToGenotypeIds[d].length;
includeExpressionSample[d] = new boolean[sampleCount];
byte[] genotypes = dSNP.getGenotypes();
for (int s = 0; s < sampleCount; s++) {
int ind = inds[s];
double valX = genotypes[ind]; // loadedSNPGenotype[ind];
if (valX != -1) {
includeExpressionSample[d][s] = true;
} else {
includeExpressionSample[d][s] = false;
}
}
} else {
dSNP.clearGenotypes();
dSNP = null;
wp.getFlipSNPAlleles()[d] = null;
snps[d] = null;
}
}
}
if (cisOnly) {
dsResults = new Result(m_numDatasets, wp.getProbes().length, wp.getId());
for (int d = 0; d < m_numDatasets; d++) {
SNP dSNP = snps[d];
if (dSNP != null) {
dsResults.numSamples[d] = snpmeancorrectedgenotypes[d].length;
double[][] rawData = m_expressiondata[d].getMatrix();
double[] varY = m_expressiondata[d].getProbeVariance();
double[] meanY = m_expressiondata[d].getProbeMean();
int samplecount = m_expressiondata[d].getIndividuals().length;
double[][] covariates = null;
if (m_covariates != null) {
DoubleMatrixDataset<String, String> covariateData = m_covariates[d];
covariates = covariateData.rawData;
}
for (int p = 0; p < probes.length; p++) {
int pid = probes[p];
Integer probeId = m_probeTranslation.get(d, pid);
if (probeId != -9) {
test(
d,
p,
probeId,
snpmeancorrectedgenotypes[d],
originalgenotypes[d],
snpvariances[d],
varY[probeId],
meanY[probeId],
includeExpressionSample[d],
samplecount,
rawData,
covariates,
dsResults,
this.currentWP,
this.metaAnalyseModelCorrelationYHat,
this.metaAnalyseInteractionTerms,
this.determinefoldchange);
} else {
dsResults.correlations[d][p] = Double.NaN;
dsResults.zscores[d][p] = Double.NaN;
}
}
} else {
for (int p = 0; p < probes.length; p++) {
dsResults.correlations[d][p] = Double.NaN;
dsResults.zscores[d][p] = Double.NaN;
}
}
}
} else if (transOnly) {
HashSet<Integer> probestoExclude = null;
if (probes != null) {
probestoExclude = new HashSet<Integer>();
for (int p = 0; p < probes.length; p++) {
probestoExclude.add(probes[p]);
}
}
dsResults = new Result(m_numDatasets, m_numProbes, wp.getId());
for (int d = 0; d < m_numDatasets; d++) {
SNP dSNP = snps[d];
dsResults.numSamples[d] = snpmeancorrectedgenotypes[d].length;
double[][] rawData = m_expressiondata[d].getMatrix();
double[] varY = m_expressiondata[d].getProbeVariance();
double[] meanY = m_expressiondata[d].getProbeMean();
int samplecount = m_expressiondata[d].getIndividuals().length;
if (dSNP != null) {
dsResults.numSamples[d] = snpmeancorrectedgenotypes[d].length;
for (int pid = 0; pid < m_numProbes; pid++) {
if (probestoExclude == null || !probestoExclude.contains(pid)) {
Integer probeId = m_probeTranslation.get(d, pid);
if (probeId != -9) {
test(
d,
pid,
probeId,
snpmeancorrectedgenotypes[d],
originalgenotypes[d],
snpvariances[d],
varY[probeId],
meanY[probeId],
includeExpressionSample[d],
samplecount,
rawData,
null,
dsResults,
this.currentWP,
this.metaAnalyseModelCorrelationYHat,
this.metaAnalyseInteractionTerms,
this.determinefoldchange);
} else {
dsResults.correlations[d][pid] = Double.NaN;
dsResults.zscores[d][pid] = Double.NaN;
}
} else {
dsResults.correlations[d][pid] = Double.NaN;
dsResults.zscores[d][pid] = Double.NaN;
}
}
} else {
for (int p = 0; p < m_numProbes; p++) {
dsResults.correlations[d][p] = Double.NaN;
dsResults.zscores[d][p] = Double.NaN;
}
}
}
} else {
dsResults = new Result(m_numDatasets, m_numProbes, wp.getId());
for (int d = 0; d < m_numDatasets; d++) {
SNP dSNP = snps[d];
dsResults.numSamples[d] = snpmeancorrectedgenotypes[d].length;
double[][] rawData = m_expressiondata[d].getMatrix();
double[] varY = m_expressiondata[d].getProbeVariance();
double[] meanY = m_expressiondata[d].getProbeMean();
int samplecount = m_expressiondata[d].getIndividuals().length;
if (dSNP != null) {
dsResults.numSamples[d] = snpmeancorrectedgenotypes[d].length;
// RunTimer t2 = new RunTimer();
for (int pid = 0; pid < m_numProbes; pid++) {
Integer probeId = m_probeTranslation.get(d, pid);
if (probeId != -9) {
test(
d,
pid,
probeId,
snpmeancorrectedgenotypes[d],
originalgenotypes[d],
snpvariances[d],
varY[probeId],
meanY[probeId],
includeExpressionSample[d],
samplecount,
rawData,
null,
dsResults,
this.currentWP,
this.metaAnalyseModelCorrelationYHat,
this.metaAnalyseInteractionTerms,
this.determinefoldchange);
} else {
dsResults.correlations[d][pid] = Double.NaN;
dsResults.zscores[d][pid] = Double.NaN;
}
}
// System.out.println("Test: "+t2.getTimeDesc());
} else {
for (int p = 0; p < m_numProbes; p++) {
dsResults.correlations[d][p] = Double.NaN;
dsResults.zscores[d][p] = Double.NaN;
}
}
}
}
convertResultsToPValues(wp, dsResults);
if (m_eQTLPlotter != null) {
for (int p = 0; p < dsResults.pvalues.length; p++) {
double pval = dsResults.pvalues[p];
if (!Double.isNaN(pval)) {
if (pval < m_pvaluePlotThreshold) {
ploteQTL(wp, p);
}
}
}
}
snps = wp.getSnps();
if (snps != null) {
for (SNP snp : snps) {
if (snp != null) {
snp.clearGenotypes();
}
}
}
// if result output is binary, convert to bytes and deflate the set of bytes.
// if (m_binaryoutput) {
// deflateResults(wp);
// }
// now push the results in the queue..
try {
wp.setNumTested(testsPerformed);
m_result_queue.put(wp);
} catch (InterruptedException e) {
e.printStackTrace();
}
// System.out.println("Analyze: "+t1.getTimeDesc());
}
