Recombinant protein production in microbial systems is a widely used technique, yet up to half of these experiments fail at the expression phase. A number of contributing factors to failures have been proposed, e.g. codon-bias, mRNA folding, mRNA:ncRNA avoidance, tRNA abundance and G+C content. Determining which, if any, of these features explains experiment failures is an active area of research. We have discovered that an ensemble energy model of RNA folding that captures the accessibility of translation initiation sites greatly outperforms other features in predicting the outcomes of 11,430 recombinant protein expression experiments in Escherichia coli. We have developed a new computational tool called Tisigner, that optimises the first nine codons of an mRNA to improve (or impair) accessibility. Our evaluations have shown that this approach is generally sufficient to elevate the chances of a successful experiment, with the advantage that straightforward PCR cloning methods can be used to integrate optimised sequences.