In JeffreyMolendijk/ml-viewR: Machine learning viewer

knitr::opts_chunk$set(
  dev = 'svg',
  collapse = TRUE,
  comment = "#>",
  fig.width = 6,
  fig.height = 4
)

Introduction

ml-viewR is a visualization package for machine learning analyses. The functions in mlview are named according to the package which they are to be used with, whereas generic functions start with "mlview.".

ropls

This package contains functions used for visualization of ropls objects. mlview currently supports all models created by ropls, including PCA, PLS, OPLS, PLS-DA and OPLS-DA models.

The ropls.plot function can be used to create score (default) and loading plots. This function allows colour by group and the addition of group ellipses. Hotelling's T2 ellipse can be toggled on or off. The ggplot output from ropls.plot can be further modified if required.

The ropls.enrich function performs fgsea enrichment on a matrix of variables, variable information and a single numeric comlumn. Furthermore, the column(s) used in the enrichment analysis can be specified. This function returns the fgsea output, including enriched groups, enrichment scores and p-values.

mlviewer for ropls models

The functions in mlviewer starting with "ropls" (e.g. ropls.plot, ropls.enrich) can be used to visualize the results from models created using the package ropls.

Installation

To install mlviewer from GitHub run the following code:

library(devtools)   
install_github("JeffreyMolendijk/ml-viewR")

Analysis example

Required libraries

To get started we need to load some packages.

library(mlviewer)
library(ropls)
library(ggplot2)
library(plotly)
library(dplyr)
library(fgsea)
library(rlang)
library(shiny)

Creating unsupervised and supervised models

The mtcars dataset is used for this example. We will create a PCA, OPLS and OPLS-DA model using ropls. To create the PCA model we supply the data matrix, but no y-variable. For the OPLS model we supply a continuous variable ("mpg"), and set the number of predictive and orthogonal components to 1. For the OPLS-DA model we supply a categorical variable with two groups ("am"). Regular PLS and PLS-DA models can be made by setting the number of orthogonal components to 0.

#Load mtcars dataset
data.x = mtcars

#Make models
ropls.pca = opls(x = mtcars, predI = 2, info.txtC = NULL, fig.pdfC = FALSE)
ropls.opls = opls(x = mtcars %>% select(-mpg), y = mtcars$mpg, predI = 1, orthoI = 1, info.txtC = NULL, fig.pdfC = FALSE)
ropls.oplsda = ropls1 = opls(x = mtcars %>% select(-am), y = as.factor(mtcars$am), predI = 1, orthoI = 1, info.txtC = NULL, fig.pdfC = FALSE)

Visualizing results

Produce ggplot images of ropls models using ropls.plot.

ropls.plot(ropls.pca, xvar = "p1", yvar = "p2", hotelling = TRUE, ellipse = FALSE)
ropls.plot(ropls.pca, xvar = "p1", yvar = "p2", hotelling = TRUE, ellipse = FALSE, col.pca = data.x$mpg)
ropls.plot(ropls.pca, xvar = "p1", yvar = "p2", hotelling = TRUE, ellipse = FALSE, col.pca = as.factor(data.x$am))

ropls.plot(ropls.pca, plottype = "loading", xvar = "p1", yvar = "p2", hotelling = TRUE, ellipse = FALSE)

ropls.plot(ropls.opls, xvar = "p1", yvar = "o1", hotelling = TRUE)
ropls.plot(ropls.opls, xvar = "p1", yvar = "o1", hotelling = TRUE, plottype = "loading")

ropls.plot(ropls.oplsda, xvar = "p1", yvar = "o1", hotelling = TRUE)
ropls.plot(ropls.oplsda, xvar = "p1", yvar = "o1", hotelling = TRUE, ellipse = TRUE) 

ropls.plot(ropls.oplsda, xvar = "p1", yvar = "o1", hotelling = TRUE, ellipse = TRUE) %>% ggplotly()

Produce interactive plots

Combining mlviewer with plotly allows the user to study the individual scores and loadings more easily.

ropls.plot(ropls.oplsda, xvar = "p1", yvar = "o1", hotelling = TRUE, ellipse = TRUE) %>% ggplotly()

Shiny results

# basic example
shinyApp(
  ui = fluidPage(
    selectInput("plot", "Plot:",
                c("PCA, scoreplot, no colour" = "p1",
                  "PCA, scoreplot, continuous colour" = "p2",
                  "PCA, scoreplot, categorical colour" = "p3")),
    plotOutput('plot')
  ),
  server = function(input, output) {

    output$plot <- renderPlot({
      if (input$plot == "p1") {
      ropls.plot(ropls.pca, xvar = "p1", yvar = "p2", hotelling = TRUE, ellipse = FALSE)
    } else if (input$plot == "p2") {
      ropls.plot(ropls.pca, xvar = "p1", yvar = "p2", hotelling = TRUE, ellipse = FALSE, col.pca = data.x$mpg)
    } else if (input$plot == "p3") {
      ropls.plot(ropls.pca, xvar = "p1", yvar = "p2", hotelling = TRUE, ellipse = FALSE, col.pca = as.factor(data.x$am))
    }
    })

  }
)