Massively parallel installation and evaluation of cancer coding mutations

Most cancer coding mutations are of unknown function, limiting their biological and therapeutic interpretation. While prime editing enables precise genomic alterations, mutations often reside in sequence contexts unfavorable to current prime-editing guide RNA (pegRNA) designs. Here we develop CodonPrime, a prime-editing framework that exploits codon degeneracy surrounding a mutation site to substantially expand its targeting set of pegRNAs. We use this approach to screen ~2500 coding mutations spanning 298 cancer genes, yielding a 9.2-fold increase in amino-acid editing efficiency over conventional prime editing. Approximately 10% of coding mutations enhance cell growth, recovering known oncogenic hotspots and revealing oncogenic potential for previously uncharacterized mutations (e.g. HNF1A R272C ). Our analysis uncovers dominant effects in paralogous genes (e.g. RHOBTB1/2) and organization of mutant phenotypes in pathways (e.g. PI3K-MTOR signaling). We formulate a general-purpose transformer for design of CodonPrime pegRNAs, enabling scalable interrogation of human coding variants.