Variable selection

imbs-hl · Jul 17, 2024 · 5219d54 · 5219d54
1 parent 0b48838
commit 5219d54
Show file tree

Hide file tree

Showing 2 changed files with 161 additions and 2 deletions.
diff --git a/README.Rmd b/README.Rmd
@@ -38,7 +38,7 @@ devtools::install_github("imbs-hl/fuseMLR")
 
 ### Usage example
 
-The following example is based on simulated data available in `fuseMLR`. Data have been simulated using the R package `InterSIM`, version 2.2.0.
+The following example is based on simulated data available in `fuseMLR`. Data have been simulated using the R package `InterSIM`, version 2.2.0. 
 
 - Let us inspect our simulated data.
 
@@ -50,6 +50,12 @@ data("entities")
 str(object = entities, max.level = 2)
 ```
 
+Variable selection, training and prediction are the main functionalities of `fuseMLR`. As variable selection and training are performed for a training study, predictions are made for a new study.
+
+#### A) Preparation of a training study
+
+We need to set up a study, its layers and the training data entities.
+
 - Instantiate a training study: A study is the fundamental component of a `fuseMLR` object.
 
 ```{r training_study, include=TRUE, eval=TRUE}
@@ -92,6 +98,49 @@ print(train_study)
 
 - An upset plot of the training study: Visualize patient overlap across layers.
 
-```{r upsetplot, include=TRUE, eval=TRUE}
+```{r upsetplot, include=TRUE, eval=TRUE, }
 train_study$upset(order.by = "freq")
 ```
+
+#### B) Variable selection
+
+We need to set up variable selection methods to our training study. Note that this can be the same method or different layer-specific methods. For simplicity, we will set up the same method on all layers.
+
+- Preparation parameters of the variable selection method.
+
+```{r varsel_param, include=TRUE, eval=TRUE}
+same_param_varsel <- ParamVarSel$new(id = "ParamVarSel",
+                                     param_list = list(num.trees = 1000, mtry = 3))
+print(same_param_varsel)
+```
+
+- Instantiate the variable selection method and assign training layers.
+
+```{r varsel_object, include=TRUE, eval=TRUE}
+varsel_ge <- VarSel$new(id = "varsel_geneexpr",
+                              package = "Boruta",
+                              varsel_fct = "Boruta",
+                              param = same_param_varsel,
+                              train_layer = tl_ge)
+
+varsel_pr <- VarSel$new(id = "varsel_geneexpr",
+                                 package = "Boruta",
+                                 varsel_fct = "Boruta",
+                                 param = same_param_varsel,
+                                 train_layer = tl_pr)
+
+varsel_me <- VarSel$new(id = "varsel_geneexpr",
+                                 package = "Boruta",
+                                 varsel_fct = "Boruta",
+                                 param = same_param_varsel,
+                                 train_layer = tl_me)
+```
+
+- Perform variable selection on our training study.
+
+```{r varsel, include=TRUE, eval=TRUE}
+var_sel_res <- train_study$varSelection()
+print(var_sel_res)
+```
+
+For each layer the variable selection results show which variable have been selected.
diff --git a/README.md b/README.md
@@ -81,6 +81,14 @@ str(object = entities, max.level = 2)
     ##   ..$ proteinexpr:'data.frame':  23 obs. of  162 variables:
     ##   ..$ methylation:'data.frame':  23 obs. of  369 variables:
 
+Variable selection, training and prediction are the main functionalities
+of `fuseMLR`. As variable selection and training are performed for a
+training study, predictions are made for a new study.
+
+#### A) Preparation of a training study
+
+We need to set up a study, its layers and the training data entities.
+
 - Instantiate a training study: A study is the fundamental component of
   a `fuseMLR` object.
 
@@ -163,3 +171,105 @@ train_study$upset(order.by = "freq")
 ```
 
 ![](README_files/figure-gfm/upsetplot-1.png)<!-- -->
+
+#### B) Variable selection
+
+We need to set up variable selection methods to our training study. Note
+that this can be the same method or different layer-specific methods.
+For simplicity, we will set up the same method on all layers.
+
+- Preparation parameters of the variable selection method.
+
+``` r
+same_param_varsel <- ParamVarSel$new(id = "ParamVarSel",
+                                     param_list = list(num.trees = 1000, mtry = 3))
+print(same_param_varsel)
+```
+
+    ## Class: ParamVarSel
+    ## id   : ParamVarSel
+    ## Parameter combination
+    ## $num.trees
+    ## [1] 1000
+    ## 
+    ## $mtry
+    ## [1] 3
+
+- Instantiate the variable selection method and assign training layers.
+
+``` r
+varsel_ge <- VarSel$new(id = "varsel_geneexpr",
+                              package = "Boruta",
+                              varsel_fct = "Boruta",
+                              param = same_param_varsel,
+                              train_layer = tl_ge)
+
+varsel_pr <- VarSel$new(id = "varsel_geneexpr",
+                                 package = "Boruta",
+                                 varsel_fct = "Boruta",
+                                 param = same_param_varsel,
+                                 train_layer = tl_pr)
+
+varsel_me <- VarSel$new(id = "varsel_geneexpr",
+                                 package = "Boruta",
+                                 varsel_fct = "Boruta",
+                                 param = same_param_varsel,
+                                 train_layer = tl_me)
+```
+
+- Perform variable selection on our training study.
+
+``` r
+var_sel_res <- train_study$varSelection()
+print(var_sel_res)
+```
+
+    ##          Layer        variable
+    ## 1     geneexpr           ACACA
+    ## 2     geneexpr            BAP1
+    ## 3     geneexpr            CDH3
+    ## 4     geneexpr           CHEK2
+    ## 5     geneexpr           EIF4E
+    ## 6     geneexpr          MAP2K1
+    ## 7     geneexpr          MAPK14
+    ## 8     geneexpr            PCNA
+    ## 9     geneexpr           YWHAE
+    ## 10    geneexpr           YWHAZ
+    ## 11 proteinexpr        Bap1.c.4
+    ## 12 proteinexpr             Bid
+    ## 13 proteinexpr       Cyclin_E2
+    ## 14 proteinexpr      P.Cadherin
+    ## 15 proteinexpr            Chk1
+    ## 16 proteinexpr      Chk1_pS345
+    ## 17 proteinexpr            EGFR
+    ## 18 proteinexpr     EGFR_pY1173
+    ## 19 proteinexpr     HER3_pY1289
+    ## 20 proteinexpr           MIG.6
+    ## 21 proteinexpr           ETS.1
+    ## 22 proteinexpr MEK1_pS217_S221
+    ## 23 proteinexpr        p38_MAPK
+    ## 24 proteinexpr    c.Met_pY1235
+    ## 25 proteinexpr           N.Ras
+    ## 26 proteinexpr            PCNA
+    ## 27 proteinexpr     PEA15_pS116
+    ## 28 proteinexpr PKC.delta_pS664
+    ## 29 proteinexpr           Rad50
+    ## 30 proteinexpr     C.Raf_pS338
+    ## 31 proteinexpr          p70S6K
+    ## 32 proteinexpr    p70S6K_pT389
+    ## 33 proteinexpr           Smad4
+    ## 34 proteinexpr     STAT3_pY705
+    ## 35 proteinexpr  14.3.3_epsilon
+    ## 36 methylation      cg20139214
+    ## 37 methylation      cg18457775
+    ## 38 methylation      cg01306510
+    ## 39 methylation      cg02412050
+    ## 40 methylation      cg07566050
+    ## 41 methylation      cg02630105
+    ## 42 methylation      cg20849549
+    ## 43 methylation      cg00547829
+    ## 44 methylation      cg25539131
+    ## 45 methylation      cg07064406
+
+For each layer the variable selection results show which variable have
+been selected.