Ineffective heterologous protein synthesis has often been ascribed to codon bias and rare codons. New experimental evidence suggests that codon bias alone may not be the sole cause of poor translation. In this paper we present a free-energy based model of translation elongation to predict and optimize genes for expression in E. coli. The model takes into account second order free energy effects from the binding between the anti-Shine-Dalgarno sequence of the 3' terminal 16S rRNA tail and the mRNA, tRNA abundance, and ribosome displacement. The model and software allow optimization of genes for increased (or decreased) protein yield. The model's predictive and optimization accuracy was assessed by optimizing and expressing three model genes and multiple mRNA variants coding for GST (26 kDa Glutathion S-Transferase from Schistosoma japonicum). Protein yield of optimized genes showed increase from their wildtype levels. Optimization of Glutathion S-Transferase from Schistosoma japonicum and Alcohol Dehydrogenase from Clostridium ljungdahlii DSM 13528 are discussed as examples.