R/makeGrittyGuess.R

In RadioGx: Analysis of Large-Scale Radio-Genomic Data

# Generate initial model parameters for fitting linear models
#
# @return Estimated parameters for first model fit
#
.makeGrittyGuess <- function(lower_bounds, upper_bounds, D, SF) {
    dFromD2 <- abs(D - 2)
    dFromSF10 <- abs(exp(SF) - 0.1)
    D10Ind <- which.min(dFromSF10)[1]
    SF2Ind <- which.min(dFromD2)[1]

    if (D10Ind == SF2Ind) {
        dFromD2[SF2Ind] <- dFromD2[SF2Ind] + dFromD2[1] #make it not the smallest anymore
        SF2Ind <- which.min(dFromD2)[1]
    }

    DSF2 <- D[SF2Ind]
    SF2 <- SF[SF2Ind]
    D10 <- D[D10Ind]
    SFD10 <- SF[D10Ind]

    if(any(c(DSF2, D10, D[D10Ind] - D[SF2Ind])==0)){
        return(c(0.25, 0.25))
    }   
    gritty_guess <- pmin(pmax(c((SF[D10Ind] * D[SF2Ind] ^ 2 - SF[SF2Ind] * D[D10Ind] ^ 2) / D[SF2Ind] / D[D10Ind] / (D[D10Ind] - D[SF2Ind]),
                                (SF[D10Ind] * D[SF2Ind] - SF[SF2Ind] * D[D10Ind]) / D[SF2Ind] / D[D10Ind] / (D[SF2Ind] - D[D10Ind])), 
                                lower_bounds), 
                                upper_bounds) # assumes the SF2Indth point is SF2 and D10Indth point is D10 and imputes alpha, 
                                              # beta from that assumption unless either would thus be out of bounds 
    return(gritty_guess)
}