Fig. 1: Massively Parallel Reporter Assays (MPRAs) to measure cell type-specific 5’UTR regulation of translation.

A A model-based design strategy for 5’UTRs in mRNA therapeutics applications, using neural network-based predictive models trained on MRPA data. B Summary of polysome profiling MPRA. A library with a randomized 50nt 5’UTR region was synthesized as in vitro transcribed (IVT) mRNA, transfected into HEK293T, T cells, and HepG2 cells, and fractionated using a sucrose gradient to separate mRNAs with different numbers of ribosomes. Fractions were then barcoded and sequenced, and the Mean Ribosome Load (MRL) was calculated for each 5’UTR variant as a proxy of translation efficiency. The resulting data contained 204,803 5’UTR variants with 100 or more reads in all replicates, in two replicates in HEK293T, two in T cells, and one in HepG2. C Comparison of MRL measurements across cell lines. 5’UTR variants were sorted by the minimum number of reads across all replicates, and the top 20,000 were used for this analysis. Data from only one replicate per cell line is shown. Data from additional replicates can be found in Supplementary Fig. 1. D Architecture of Optimus 5-Prime, a convolutional neural network model for predicting MRL from 5’UTR sequence³³. E Optimus 5-Prime predictions compared to MRL measurements in all three cell lines. The top 20,000 5’UTRs by read count in HEK293, which were not used for model training, were used for this analysis. Source data are provided as a Source Data file.