A machine learning approach identifies principles and determinants of eukaryotic ribosome pausing

The regulation of translation elongation rate is a fundamental problem in biology. To identify principles governing eukaryotic ribosome pausing we integrated unsupervised machine learning and ribosome profiling. We find tRNA abundance drives incorporation of hydrophobic amino acids into the ribosome active site but non-hydrophobic residues are accepted less efficiently regardless of (tRNA abundance and?) codon optimality. Altering tRNA levels demonstrates codon decoding via wobble pairing is slower than decoding by cognate tRNA; surprisingly it also uncovers that even rare tRNAs engage in widespread wobble interactions with near-cognate codons leading to ribosomal pausing. Ribosome Thus, codon choice and tRNA pools must be carefully tuned to harmonize elongation rate with cotranslational polypeptide maturation, while minimize unwanted pausing leading to collisions induced by apposition of fast upstream and slow downstream ribosomes.