Optimization of gene sequences for protein expression

A method for tailoring gene transcript sequences for protein expression, comprising steps:

• a) measuring ribosome dynamics in an organism or cell type of interest to obtain ribosome profiling data;
• b) training a statistical model of the relationship between DNA sequence and translation speed on the ribosome profiling data;
• and
• c) using the trained model to design a DNA sequence encoding and tailored for expression of a protein of interest,
• wherein step (b) comprises training the model to learn sequence preferences for translation by using counts of fragments at each codon position to learn the cell type or organism's sequence preferences for fast translation, by:
• counting how many ribosomes are seen at each codon position in each gene transcript, normalized by the average number of ribosomes per position in that gene transcript;
• using a machine learning protocol to learn the model for the position-to-normalized count relationship of the genome, wherein a 30-40 nucleotide window is encoded around each codon feature;
• and
• training a neural network to predict normalized counts as a function of these features, providing the model that can take as input any arbitrary gene transcript sequence and predict how ribosomes will slow down or speed up on that sequence.