tumor only module improvement

Author: Beifang

README.md

@@ -1,9 +1,11 @@
 MSIsensor
 ===========
-MSIsensor is a C++ program to detect replication slippage variants at microsatellite regions, and differentiate them as somatic or germline. Given paired tumor and normal sequence data, it builds a distribution for expected (normal) and observed (tumor) lengths of repeated sequence per microsatellite, and compares them using Pearson's Chi-Squared Test. Comprehensive testing indicates MSIsensor is an efficient and effective tool for deriving MSI status from standard tumor-normal paired sequence data. Since there are many users complained that they don't have paired normal sequence data or related normal sequence data can be used to build a paired normal control, we released MSIsensor with version from 0.3. Given tumor only sequence data, it uses comentropy theory and figures out a comentropy value for a distribution per microsatellite. Our test results show that it's performance is comparable with paired tumor and normal sequence data input(figure below). We suggest msi score cutoff 11% for tumor only data. (msi high: msi score >= 11%).
+MSIsensor is a C++ program to detect replication slippage variants at microsatellite regions, and differentiate them as somatic or germline. Given paired tumor and normal sequence data, it builds a distribution for expected (normal) and observed (tumor) lengths of repeated sequence per microsatellite, and compares them using Pearson's Chi-Squared Test. Comprehensive testing indicates MSIsensor is an efficient and effective tool for deriving MSI status from standard tumor-normal paired sequence data. Since there are many users complained that they don't have paired normal sequence data or related normal sequence data can be used to build a paired normal control, we released MSIsensor with version from 0.3. Given tumor only sequence data, it uses comentropy theory and figures out a comentropy value for a distribution per microsatellite. Our test results show that it's performance is comparable with paired tumor and normal sequence data input(figure below). And, We recommend to set different msi score cutoff values for different cancer types. (for example: TCGA UCEC, msi high: msi score >= 13%). We also provide test result of TCGA and EGA data.(see AUC figures below)
 
 ![](https://github.com/ding-lab/msisensor/blob/master/test/tumor_only_vs_pair.jpg)
 
+![](https://github.com/ding-lab/msisensor/blob/master/test/msisensor-tumor-only.png)
+
 If you used this tool for your work, please cite [PMID 24371154](https://www.ncbi.nlm.nih.gov/pubmed/24371154)
 
 Beifang Niu*, Kai Ye*, Qunyuan Zhang, Charles Lu, Mingchao Xie, Michael D. McLellan, Michael C. Wendl and Li Ding#.MSIsensor: microsatellite instability detection using paired tu-mor-normal sequence data. Bioinformatics 30, 1015–1016 (2014).

homo.cpp

@@ -316,7 +316,8 @@ void HomoSite::DisTumorSomatic(Sample &sample) {
         withSufCov = false;
         comentropy = 0;
     }
-    if (comentropy >= paramd.comentropyThreshold) { 
+    //if (comentropy >= paramd.comentropyThreshold) {
+    if (withSufCov and comentropy >= paramd.comentropyThreshold) { 
         reportSomatic = true;
         sample.numberOfMsiDataPoints ++;
     }
@@ -416,6 +417,7 @@ double HomoSite::DistanceBetweenTwo( unsigned short * FirstOriginal, unsigned sh
 double HomoSite::Comentropy( unsigned short * tumorDis, unsigned int dispots ) {
     double sum = 0;
     double comentropy = 0.0;
+    double number = 0.0;
     for (int i = 0; i < dispots; i++){
         if (tumorDis[i] <3){
             tumorDis[i] = 0;
@@ -425,10 +427,12 @@ double HomoSite::Comentropy( unsigned short * tumorDis, unsigned int dispots ) {
     if ( sum != 0 ) {
         for( int j = 0; j < dispots; j++){
             if( tumorDis[j] != 0 ){
+                number += 1;
                 comentropy -= (tumorDis[j]/sum)*log(tumorDis[j]/sum);
             }
         }
     }
+    if ( number != 0 ) comentropy = comentropy/number;
     return comentropy;
 }
 

param.cpp

@@ -103,7 +103,7 @@ Param::Param()
     , covCutoff( 20 )
     , Normalization(0)
     , fdrThreshold( 0.05 ) 
-    , comentropyThreshold( 1 ) // for tumor only
+    , comentropyThreshold( 0.3 ) // for tumor only
 {
     inital_homo_phabet();
     initalphabet();