R/two.ways.stepback.R

In maSigPro: Significant Gene Expression Profile Differences in Time Course Gene Expression Data

"two.ways.stepback" <- 
function (y = y, d = d, alfa = 0.05 , family = gaussian() , epsilon=0.00001) 
{
    OUT <- NULL
    lm1 <- glm(y ~ ., data = d, family = family, epsilon=epsilon)
    result <- summary(lm1)$coefficients[, 4]
    max <- max(result[-1], na.rm = TRUE)
    d <- d[, names(result)[-1]]
    while (max > alfa) {
        varout <- names(result)[result == max]
        pos <- position(matrix = d, vari = varout)
        OUT <- as.data.frame(cbind(OUT, d[, pos]))
        x <- ncol(OUT)
        colnames(OUT)[x] <- colnames(d)[pos]
        if (ncol(d) == 2) {
            min <- min(result[-1], na.rm = TRUE)
            lastname <- names(result)[result == min]
        }
        d <- d[, -pos]
        if (is.null(dim(d))) {
            d <- as.data.frame(d)
            colnames(d) <- lastname
        }
        j = ncol(d) + 1
        pval <- NULL
        for (i in 1:ncol(OUT)) {
            sub <- cbind(d, OUT[, i])
            sub <- as.data.frame(sub)
            lm2 <- glm(y ~ ., data = sub, family = family, epsilon=epsilon)
            result <- summary(lm2)$coefficients[, 4]
            pval[i] <- result[j + 1]
        }
        min <- min(pval, na.rm = TRUE)
        while (min <= alfa) {
            b <- pval == min
            c <- c(1:length(pval))
            pos <- c[b]
            d <- cbind(d, OUT[, pos])
            d <- as.data.frame(d)
            colnames(d)[j] <- colnames(OUT)[pos]
            if (ncol(OUT) == 2) {
                max <- max(pval, na.rm = TRUE)
                b <- pval == max
                c <- c(1:length(pval))
                last <- c[b]
                lastname <- colnames(OUT)[last]
            }
            OUT <- OUT[, -pos]
            if (is.null(dim(OUT))) {
                OUT <- as.data.frame(OUT)
                colnames(OUT) <- lastname
            }
            j = ncol(d) + 1
            pval <- NULL
            for (i in 1:ncol(OUT)) {
                sub <- cbind(d, OUT[, i])
                sub <- as.data.frame(sub)
                lm2 <- glm(y ~ ., data = sub, family = family, epsilon=epsilon)
                result <- summary(lm2)
                pval[i] <- result$coefficients[, 4][j + 1]
            }
            min <- min(pval, na.rm = TRUE)
            if (ncol(OUT) == 1) {
                if (min <= alfa) {
                  d <- cbind(d, OUT[, 1])
                  d <- as.data.frame(d)
                  colnames(d)[j] <- colnames(OUT)[1]
                }
                min = 1
            }
        }
        lm1 <- glm(y ~ ., data = d, family = family, epsilon=epsilon)
        result <- summary(lm1)$coefficients[, 4]
        max <- max(result[-1], na.rm = TRUE)
        if (length(result[-1]) == 1) {
            max <- result[-1]
            if (max > alfa) {
                max = 0
                lm1 <- glm(y ~ 1, family = family, epsilon=epsilon)
            }
        }
    }
    return(lm1)
}