optimize {permubiome}

Optimization for detection of features with larger variation between classes

Description

This function is the previous step to the permutation test and it optimizes the detection of features differentially distributed between classes. The intra- and inter-classes distances are calculated with log-transformed data and then a ratio test is done to maximize the variation between classes. This pre-processing penalizes the microbiome features with a larger intra-class and lower inter-class variation, which would interfere with the statistical estimations to executed during the permutation test.

Usage

optimize()

Author(s)

Alfonso Benitez-Paez

References

Benitez-Paez A. 2023. Permubiome: an R package to perform permutation based test for biomarker discovery in microbiome analyses. [https://cran.r-project.org]. Benitez-Paez A, et al. mSystems. 2020;5:e00857-19. doi: 10.1128/mSystems.00857-19.

Examples

## The function is currently defined as
function () 
{
    load("permubiome.RData")
    df_norm <- df_norm
    REFERENCE <- REFERENCE
    classes <- levels(as.factor(df_norm$Class))
    if (REFERENCE == "") {
        REFERENCE <- classes[1]
    }
    else if (REFERENCE == classes[2]) {
        classes[2] <- classes[1]
        classes[1] <- REFERENCE
    }
    df_norm$Class <- factor(df$Class, levels = (c(classes[1], 
        classes[2])))
    group1 <- subset(df_norm, Class == classes[1])
    group2 <- subset(df_norm, Class == classes[2])
    categories <- colnames(df_norm)
    distance_matrix <- matrix(, nrow = ncol(df_norm) - 2, ncol = 7, 
        byrow = T)
    colnames(distance_matrix) <- c("Category", paste("SumDist ", 
        "[", classes[1], "]", sep = ""), paste("SumDist ", "[", 
        classes[2], "]", sep = ""), "ClassDist", "RatioDist", 
        "Delta-Log", "Z-score")
    for (i in 3:(ncol(group1))) {
        mydata1 <- group1[, i]
        sumdist1 <- log10(sum(abs(apply(combn(mydata1, 2), 2, 
            diff))))
        distance_matrix[(i - 2), 1] <- categories[i]
        distance_matrix[(i - 2), 2] <- sumdist1
    }
    for (j in 3:(ncol(group2))) {
        mydata2 <- group2[, j]
        sumdist2 <- log10(sum(abs(apply(combn(mydata2, 2), 2, 
            diff))))
        distance_matrix[(j - 2), 3] <- sumdist2
    }
    classes_matrix <- matrix(, nrow = ncol(group1) - 2, ncol = (nrow(group1) * 
        nrow(group2)), byrow = T)
    rownames(classes_matrix) <- colnames(group1[3:ncol(group1)])
    features <- colnames(group1)
    for (k in 3:(ncol(group1))) {
        classdist <- vector()
        for (l in 1:nrow(group1[k])) {
            classdist_tmp <- as.list(abs(group2[k] - group1[l, 
                k]))
            classdist <- c(classdist, classdist_tmp[[features[k]]])
        }
        classes_matrix[(k - 2), ] <- classdist
    }
    inter.class.dist <- as.list(rowSums(classes_matrix))
    for (m in 3:(ncol(group1))) {
        distance_matrix[(m - 2), 4] <- log10(inter.class.dist[[features[m]]])
    }
    distance_matrix[, 5] <- as.numeric(distance_matrix[, 4])/((as.numeric(distance_matrix[, 
        3])/as.numeric(distance_matrix[, 2])))
    distance_matrix[, 6] <- abs(as.numeric(distance_matrix[, 
        5]) - as.numeric(distance_matrix[, 4]))
    distance_matrix[, 7] <- (as.numeric(distance_matrix[, 6]) - 
        mean(as.numeric(distance_matrix[, 6])))/sd(as.numeric(distance_matrix[, 
        6]))
    selected_features <- subset(distance_matrix, abs(as.numeric(distance_matrix[, 
        6])) > quantile(as.numeric(distance_matrix[,6]),0.96))
    save(df, df_norm, REFERENCE, classes, distance_matrix, selected_features, 
        file = "permubiome.RData")
  }