How can the medianpolish used in RMA ensure that both alpha (probe affinity) and beta (expression abundance) larger than zero? Or, RMA occasionally gives out negative expression values? Thanks! Simon