Fix Issue GenABEL-Project#21: Remove "interaction_only" option

According to Yurii's reply on a forum topic, this option never worked as intended and should have been removed long time ago: "Never use '--interaction_only' option! This is something form our early experiments with GxE, and it has proven to generate crap results (GC lambda << 1)."
PolyOmica · Mar 1, 2016 · 44284b2 · 44284b2
1 parent afb3c91
commit 44284b2
Show file tree

Hide file tree

Showing 14 changed files with 17 additions and 179 deletions.
diff --git a/doc/ChangeLog b/doc/ChangeLog
@@ -11,6 +11,8 @@
 ** Overall runtime using above settings is 5
 * Handles multiple variants of NaN (NA,Na,Nan,na,nan) correct while reading of
   mldose/mlprob files
+* Fixed Issue #21: Remove "interaction_only" option. According to Yurii's
+  post on the forum this option never worked as expected.
 * Fixed bugs #5883 & #6010: The main effect is displayed in the output with the
   `interaction_only` option. Thanks to Maksim Struchalin & Lennart Karssen for
   fixing it and to Farid Radmanesh for reporting it. This was only a display

diff --git a/doc/ProbABEL_manual.tex b/doc/ProbABEL_manual.tex
@@ -486,7 +486,6 @@ \section{Running an analysis}
          --no-head : do not report header line
          --allcov  : report estimates for all covariates (large outputs!)
          --interaction: Which covariate to use for interaction with SNP analysis (default is no interaction, 0)
-         --interaction_only: like previous but without covariate acting in interaction with SNP (default is no interaction, 0)
          --mmscore : score test in samples of related individuals. File with inverse of variance-covariance matrix (for palinear) or inverse correlation (for palogist) as input parameter
          --robust  : report robust (aka sandwich, aka Hubert-White) standard errors
          --help    : print help
@@ -604,20 +603,6 @@ \subsection{Advanced analysis options}
 variance should be fit as part of \GA{}'s
 \texttt{polygenic} procedure.
 
-The option \lstinline{--interaction_only} is like \lstinline{--interaction}
-but does not include the main effect of the covariate which is acting
-in interaction with SNP into the model. For example, if \texttt{age}
-is the second covariate in the phenotype file, using the option
-\lstinline{--interaction_only=2} leads to the following model:
-\begin{equation*}
-  \textrm{trait} \sim \textrm{sex} + \textrm{SNP} +
-  \textrm{age} \times \mathrm{SNP}.
-\end{equation*}
-
-This option is useful when
-running \lstinline{--mmscore}, in which case the main effect should
-normally estimated in the polygenic model and only the interaction
-term in the \PA{} analysis.
 
 \subsection{Running multiple analyses at once: \texttt{probabel}}
 The \texttt{bin/probabel} script is a handy wraper for the \PA{}

diff --git a/doc/pacoxph.1 b/doc/pacoxph.1
@@ -62,11 +62,6 @@ Report estimates for all covariates (large outputs!).
 .BI "\-b, \-\^\-interaction" " NUMBER"
 Which covariate to use for interaction with SNP analysis (default is no interaction, 0). \fINUMBER\fR indicates the column number of the covariate.
 .TP
-.BI "\-k, \-\^\-interaction_only" " NUMBER"
-Like
-.B \-\^\-interaction
-but excludes the covariate-only model term for the  covariate acting in interaction with the SNP (default is no interaction, 0).
-.TP
 .B \-\^\-help
 Print help.
 

diff --git a/doc/palinear.1 b/doc/palinear.1
@@ -61,11 +61,6 @@ Report estimates for all covariates (large outputs!).
 .BI "\-b, \-\^\-interaction" " NUMBER"
 Which covariate to use for interaction with SNP analysis (default is no interaction, 0). \fINUMBER\fR indicates the column number of the covariate.
 .TP
-.BI "\-k, \-\^\-interaction_only" " NUMBER"
-Like
-.B \-\^\-interaction
-but excludes the covariate-only model term for the covariate acting in interaction with the SNP (default is no interaction, 0).
-.TP
 .BI "\-v, \-\^\-mmscore" " FILE"
 Score test in samples of related individuals. The FILE argument is the name of a file with the inverse of the variance-covariance matrix.
 .TP

diff --git a/doc/palogist.1 b/doc/palogist.1
@@ -61,11 +61,6 @@ Report estimates for all covariates (large outputs!).
 .BI "\-b, \-\^\-interaction" " NUMBER"
 Which covariate to use for interaction with SNP analysis (default is no interaction, 0). \fINUMBER\fR indicates the column number of the covariate.
 .TP
-.BI "\-k, \-\^\-interaction_only" " NUMBER"
-Like
-.B \-\^\-interaction
-but excludes the covariate-only model term for the covariate acting in interaction with the SNP (default is no interaction, 0).
-.TP
 .BI "\-v, \-\^\-mmscore" " FILE"
 Score test in samples of related individuals. The FILE argument is the name of a file with the inverse correlation matrix. NOTE THAT THIS FEATURE IS STILL EXPERIMENTAL!
 .TP

diff --git a/src/command_line_settings.cpp b/src/command_line_settings.cpp
@@ -159,7 +159,7 @@ char* cmdvars::getPhefilename() const
 void cmdvars::set_variables(int argc, char * argv[])
 {
     int next_option;
-    const char * const short_options = "p:i:d:m:n:c:o:s:t:g:a:relhb:k:v:u";
+    const char * const short_options = "p:i:d:m:n:c:o:s:t:g:a:relhb:v:u";
     // b - interaction parameter
     // ADD --fv FLAG (FILEVECTOR), IN WHICH CASE USE ALTERNATIVE
     // CONSTRUCTOR FOR GENDATA
@@ -181,7 +181,6 @@ void cmdvars::set_variables(int argc, char * argv[])
     { "allcov", 0, NULL, 'l' },
     { "help", 0, NULL, 'h' },
     { "interaction", 1, NULL, 'b' },
-    { "interaction_only", 1, NULL, 'k' },
     { "mmscore", 1, NULL, 'v' },
     { "robust", 0, NULL, 'u' },
     { NULL, 0, NULL, 0 } };
@@ -252,9 +251,6 @@ void cmdvars::set_variables(int argc, char * argv[])
         case 'b':
             interaction = atoi(optarg);
             break;
-        case 'k':
-            interaction_excluded = atoi(optarg);
-            break;
         case 'v':
             inverse_filename = optarg;
             break;
@@ -273,12 +269,6 @@ void cmdvars::set_variables(int argc, char * argv[])
 }  // end of function
 
 
-bool cmdvars::isIsInteractionExcluded() const
-{
-    return is_interaction_excluded;
-}
-
-
 void cmdvars::printinfo()
 {
     print_version();
@@ -332,7 +322,6 @@ void cmdvars::printinfo()
     cout << "\t --ntraits     = "      << noutcomes << endl;
     cout << "\t --ngpreds     = "      << ngpreds << endl;
     cout << "\t --interaction = "      << interaction << endl;
-    cout << "\t --interaction_only = " << interaction_excluded << endl;
 
     if (inverse_filename != NULL)
         cout << "\t --mmscore = "      << inverse_filename << endl;
@@ -381,11 +370,6 @@ void cmdvars::printinfo()
         exit(1);
     }
 
-    if (interaction_excluded != 0)
-    {
-        interaction = interaction_excluded;  // ups
-        is_interaction_excluded = true;
-    }
     if (outfilename.compare("") == 0)
     {
         outfilename = string("regression");

diff --git a/src/command_line_settings.h b/src/command_line_settings.h
@@ -51,8 +51,6 @@ class cmdvars
     int npeople;
     int ngpreds;
     int interaction;
-    int interaction_excluded;
-    bool is_interaction_excluded;
     int robust;
     string chrom;
     string sep;
@@ -82,8 +80,6 @@ class cmdvars
         npeople = -1;
         ngpreds = 1;
         interaction = 0;
-        interaction_excluded = 0;
-        is_interaction_excluded = false; //Oh Holy Matrix, forgive me for this!
         robust = 0;
         chrom = "-1";
         str_genfilename = "";
@@ -125,7 +121,6 @@ class cmdvars
     string getStrGenfilename() const;
 
     void printinfo();
-    bool isIsInteractionExcluded() const;
 };
 
 #endif /* COMMAND_LINE_SETTINGS_H_ */
diff --git a/src/main.cpp b/src/main.cpp
@@ -140,7 +140,7 @@ int main(int argc, char * argv[])
 #if COXPH
     coxph_data nrgd = coxph_data(phd, gtd, -1);
 #else
-    regdata nrgd = regdata(phd, gtd, -1, input_var.isIsInteractionExcluded());
+    regdata nrgd = regdata(phd, gtd, -1);
 #endif
 
     std::cout << " loaded null data..." << std::flush;
@@ -175,7 +175,7 @@ int main(int argc, char * argv[])
 #if COXPH
     coxph_data rgd(phd, gtd, 0);
 #else
-    regdata rgd(phd, gtd, 0, input_var.isIsInteractionExcluded());
+    regdata rgd(phd, gtd, 0);
 #endif
     std::cout << " formed regression object...\n";
 

diff --git a/src/main_functions_dump.cpp b/src/main_functions_dump.cpp
@@ -114,7 +114,6 @@ void open_files_for_output(std::vector<std::ofstream*>& outfile,
 
 int create_phenotype(phedata& phd, const cmdvars& input_var)
 {
-    phd.set_is_interaction_excluded(input_var.isIsInteractionExcluded());
     phd.setphedata(input_var.getPhefilename(),
                    input_var.getNoutcomes(),
                    input_var.getNpeople(),

diff --git a/src/phedata.cpp b/src/phedata.cpp
@@ -48,11 +48,6 @@ phedata::phedata(const char * fname, const int noutc, const int npeople,
     setphedata(fname, noutc, npeople, interaction, iscox);
 }
 
-void phedata::set_is_interaction_excluded(const bool int_exl)
-{
-    is_interaction_excluded = int_exl;
-}
-
 
 /**
  * Read phenotype data from file.
@@ -160,12 +155,6 @@ void phedata::setphedata(const char * fname, const int noutc,
         for (int i = (2 + noutcomes); i < nphenocols; i++)
         {
             infile >> tmp;
-            if (n_model_terms == interaction && is_interaction_excluded)
-               {
-                   interaction_cov_name = tmp;
-                   n_model_terms++;
-                   continue;
-               }
 
             model = model + " + ";
             model = model + tmp;
@@ -177,27 +166,13 @@ void phedata::setphedata(const char * fname, const int noutc,
     {
         if (iscox)
         {
-           if (!is_interaction_excluded)
-           {
-               model = model + " + "
-                   + model_terms[interaction - 1]
-                   + "*SNP_A1";
-           }
-           else
-           {
-               model = model + " + " + interaction_cov_name + "*SNP_A1";
-           }
+            model = model + " + "
+                + model_terms[interaction - 1]
+                + "*SNP_A1";
         }
         else
         {
-            if (!is_interaction_excluded)
-            {
-                model = model + " + " + model_terms[interaction] + "*SNP_A1";
-            }
-            else
-            {
-                model = model + " + " + interaction_cov_name + "*SNP_A1";
-            }
+            model = model + " + " + model_terms[interaction] + "*SNP_A1";
         }
     }
     model_terms[n_model_terms++] = "SNP_A1";

diff --git a/src/phedata.h b/src/phedata.h
@@ -50,7 +50,6 @@ class phedata {
  public:
     phedata()
     {
-        is_interaction_excluded = 0;
         nids_all                = 0;
         nids                    = 0;
         noutcomes               = 0;
@@ -69,10 +68,6 @@ class phedata {
 
     void setphedata(const char * fname, const int noutc, const int npeople,
                     const int interaction, const bool iscox);
-    void set_is_interaction_excluded(const bool inter_excluded);
-
-
-    bool is_interaction_excluded;
 
     /**
      * \brief The total number of individuals in the phenotype file,