R/Jaccard_cal.R

In bioinformatist/CrossICC: An Interactive Consensus Clustering Framework for Multi-platform Data Analysis

#' Title Get fihser test p value for overlaps
#' @param vec1 a string vector
#' @param vec2 a string vector
#' @param universe total number of all strings that vec1 and vec2 comes from
#' @return  a P value
get_overlap_test_by_fisher <- function(vec1, vec2, universe) {
    mat <- matrix(c(universe - length(union(vec1, vec2)), length(setdiff(vec1, vec2)), length(setdiff(vec1, vec2)), length(intersect(vec1,
        vec2))), nrow = 2)

    fr <- fisher.test(mat, alternative = "greater")
    return(fr$p.value)
}


#' Title get jaccard index of list factors

#' @param list1 a list that contains a lot vectors
#' @param list2 a list that contains a lot vectors
#' @param universe (Optional) total number of all strings that vec1 and vec2 comes from
#' @return a data frame of Jaccard index or a list contains two dataframe (jaccard index and Fisher's test P value list )
#'
get_jarrad_index_df_fromlist <- function(list1, list2, universe = NULL) {
    # get individual jaccard index
    get_J <- function(vec1, vec2) {
        I <- length(intersect(vec1, vec2))
        S <- I/(length(vec1) + length(vec2) - I)
        return(S)
    }

    # get individual jaccard index to a list
    get_J_list <- function(vec1, list.temp) {
        res.list <- unlist(lapply(list.temp, get_J, vec1 = vec1))
        return(res.list)
    }
    # fisher P list
    get_P_list <- function(vec1, list.temp, universe) {
        P.list <- unlist(lapply(list.temp, get_overlap_test_by_fisher, vec1 = vec1, universe = universe))
        return(P.list)
    }
    # get matrix output
    res.m.list <- lapply(list1, get_J_list, list.temp = list2)
    S.df <- data.frame(matrix(unlist(res.m.list), nrow = length(list1), byrow = TRUE), stringsAsFactors = FALSE)
    colnames(S.df) <- names(list2)
    row.names(S.df) <- names(list1)
    if (!is.null(universe)) {
        P.m.list <- lapply(list1, get_P_list, list.temp = list2, universe = universe)
        p.df <- data.frame(matrix(unlist(P.m.list), nrow = length(list1), byrow = TRUE), stringsAsFactors = FALSE)
        colnames(p.df) <- names(list2)
        row.names(p.df) <- names(list1)
        return(list(J.index = S.df, P.fisher = p.df))
    } else {
        return(S.df)
    }
}
# get jaccard index of two annotated dataframe (with two column)
#' Title get jaccard index of list factors

#' @param df1 an annotated data frame with cluster at the seccond column
#' @param df2 an annotated data frame with cluster at the seccond column
#' @param universe (Optional) total number of all strings that vec1 and vec2 comes from
#' @return a data frame of Jaccard index or a list contains two dataframe (jaccard index and Fisher's test P value list )
#'
get_jarrad_index_df_fromDF <- function(df1, df2, universe = NULL) {
    if (is(df1, "integer")) {
        df1 <- data.frame(names(df1), df1)
    }
    if (is(df2, "integer")) {
        df2 <- data.frame(names(df2), df2)
    }
    # get individual jaccard index
    list1 <- tapply(df1[, 1], df1[, 2], list)
    list2 <- tapply(df2[, 1], df2[, 2], list)
    if (is.null(universe)) {
        return(get_jarrad_index_df_fromlist(list1, list2))
    } else {
        return(get_jarrad_index_df_fromlist(list1, list2, universe))
    }

}

# get Rand index
#' Title Adjust Rank Index

#' @param df input data frame
#' @param col1 name of interest variable 1 column in df
#' @param col2 name of interest variable 2 column in df
#' @return adjust ARI value
rand.index <- function(df, col1, col2) {
    group1 <- as.numeric(as.factor(df[, col1]))
    group2 <- as.numeric(as.factor(df[, col2]))

    x <- abs(vapply(group1, function(x) x - group1, numeric(1)))
    x[x > 1] <- 1
    y <- abs(vapply(group2, function(x) x - group2, numeric(1)))
    y[y > 1] <- 1
    sg <- sum(abs(x - y))/2
    bc <- choose(dim(x)[1], 2)
    ri <- 1 - sg/bc
    return(ri)
}


# get adjust Rand index
#' Title Adjust Rank Index
#' @param df input data frame
#' @param col1 name of interest variable 1 column in df
#' @param col2 name of interest variable 2 column in df
#' @return adjust ARI value
#'
Cal.ARI <- function(df, col1, col2) {
    x = df[, col1]
    y = df[, col2]

    if (length(x) != length(y)) {
        stop("two vectors have different lengths!\n")
    }

    cdsum = function(x) {
        y = x * (x - 1)/2
        return(y)
    }

    mkMatrix = table(x, y)
    mkMatrixSum = apply(mkMatrix, 1, cdsum)
    matrixSum = sum(mkMatrixSum)

    matrixColsum = colSums(mkMatrix)
    matrixRowsum = rowSums(mkMatrix)
    n = length(x)

    statRow = sum(cdsum(matrixRowsum))
    statCol = sum(cdsum(matrixColsum))

    ARI = (matrixSum - (statRow * statCol)/(cdsum(n)))/(((statRow + statCol)/2) - (statRow * statCol/cdsum(n)))

    return(ARI)
}