In bhioswego/CBRMSR: CBRMSR

knitr::opts_chunk$set(echo = TRUE)

Case Based Reasoning with Multi-Stage Retrieval

CBRMSR is Case Based Reasoning with Multi-Stage Retrieval. The package allows for splitting or folding, feature selection with a balanced iterative random forest or random KNN algorithm, and class balancing with ADASYN or SMOTE. Distance matrices can be created on confounding variable data, and numeric predictor variable data. During retrieval, cases are retrieved using a confidence metric that automaticallyretrieves cases for each test case until a confidence threshold is reached. First, cases are retrieved fromsimilar confounding attributes, followed by retrieving from similar predictor attributes.

Requirements

Three dataframes are required.

1. A dataset of predictor values
2. A dataset of confounding variables (such as confounding data) to draw samples from first.
3. A classframe, which is a dataframe with two columns. The first column is sample names and the second column is classification labels.

Loading CBRMSR

suppressMessages(library(caret))
suppressMessages(library(smotefamily))
suppressMessages(library(imbalance))
suppressMessages(library(randomForest))
suppressMessages(library(stats))
suppressMessages(library(sidier))
suppressMessages(library(R6))
suppressMessages(library(sna))
suppressMessages(library(cluster))
suppressMessages(library(tictoc))
suppressMessages(library(nomclust))
suppressMessages(library(analogue))
suppressMessages(library(data.table))
suppressMessages(library(rknn))
suppressMessages(library(plyr))

 library(CBRMSR)

Included Data

Three dataframes are included for testing purposes. Predictor is a subsetted dataset of unique probes from differentially methylated regions in the TCGA-BRCA dataset. Confounding is the associated confounding variables that have been converted to numeric. Ages were assigned to a group based on their placement in an age range. Classframe is a dataframe where the first column is the TCGA sample names, and the second column is their classification label.

data(predictor)
data(confounding)
data(classframe)

Dimensions of the included predictor data

dim(predictor)

Dimensions of the included confounding data

dim(confounding)

Dimensions of the included classframe

dim(classframe)

Quick look at the included predictor data

predictor[1:5,1:5]

Quick look at the included confounding data

confounding[1:5,1:5]

Quick look at the included classframe

head(classframe, n = 5)

Creating a CBRMSR Class object

CBRMSR <- create_CBRMSR(predictor = predictor, confounding = confounding, classframe = classframe)

Splitting or creating K-Folds

SplitPercent is the percentage of data that will be used for the training set, in decimal form

CBRMSR <- splitting_module(CBRMSR, SplitPercent = 0.75)
CBRMSR <- folding_module(CBRMSR, Folds = 10)

Feature Selection using Balanced Iterative Random Forest

CBRMSR <- selection_module(CBRMSR, method = "BIRF")

Creating Distance Modules

Calculate the categorical and numeric distances. Categorical distances can be calculated using the Goodall or the Lin algorithm.

CBRMSR <- distance_module(CBRMSR, categorical.similarity = "Goodall", confounding.type = "categorical", feature.weights = TRUE)

Classifying with the two-stage module

This module runs a two-stage process. First, it retrieves samples from similar confounding factors before reducing the pool of retrieved samples through using similarity among the predictor variables. This is done automatically using a confidence metric. Each training sample is assigned a confidence value which is the average distance to samples of a different class minus the average distance to samples of the same class. This value is normalized between 0 and 1. This is the last module that should be run.

CBRMSR <- two_stage_module(CBRMSR)

Accessors

Used with \$. Example, if "myCBRMSR" was the name of your CBRMSR object, and you wanted to access the confusion matrices for the testing data, the syntax is: myCBRMSR\$testing.confusion.matrices

• predictor
• The original predictor data
• confounding
• The original confounding data
• classframe
• The dataframe of samples and classification labels
• num
• Number of training and testing sets
• training.sets
• List of dataframes, where the dataframes are training sets
• testing.sets
• List of dataframes, where the dataframes are testing sets
• training.labels
• The true classification labels for the training data
• testing.labels
• The true classification labels for the testing data
• training.confounding.sets
• The subsetted confounding data for the training sets
• testing.confounding.sets
• The subsetted confounding data for the testing sets
• selected.features
• The remaining features after feature selection and their associated importance levels (mean decrease Gini)
• feature.weights
• The Mean Decrease Gini for the features
• balanced
• A boolean value of whether or not class balancing was applied
• balanced.training.labels
• the classification labels of the training set after balancing
• balanced.training.sets
• the training datsets after balancing
• training.confounding.distances
• the distance matrices of the training confounding data
• testing.confounding.distances
• the distance matrices of the testing confounding data
• training.predictor.distances
• the distance matrices of the training predictor data
• testing.predictor.distances
• the distance matrices of the testing predictor data
• training.confusion.matrices
• confusion matrices for the training sets
• training.predicted.labels
• predicted classification labels for the training data
• testing.confusion.matrices
• confusion matrices for the testing sets
• testing.predicted.labels
• predicted classification labels for the testing data
• retrieved.samples
• Which samples were retrieved during testing

bhioswego/CBRMSR documentation built on Dec. 6, 2020, 3:16 p.m.