GROQ-seq Function Measurements for TEV Protease ML Libraries

This dataset contains GROQ-seq functional measurements of an ML-designed TEV protease variant library generated by seven generative protein-design models — PSSM, Potts/EVCouplings, EVE, ESM2, Tranception, ESM-IF1, and ProteinMPNN — under a unified Gibbs-sampling framework, alongside an error-prone PCR (epPCR) baseline. Variants are organized into three section libraries (residues 1–78, 79–157, 158–236) and a fully combinatorial library, each filtered for a complete catalytic triad and ≤10 mutations from wild type. Fitness is measured via the same split-DHFR system as the TEV protease pilot across 24 experimental conditions independently varying TEV expression, split-DHFR expression, and trimethoprim selection. This run of the assay was performed at the Living Measurements Systems Foundry (LMSF) at the National Institute of Standards and Technology, alongside the TEV protease pilot. Part of a broader effort to benchmark generative protein-design models against experimental enzyme function — providing a standardized evaluation framework for sequence-, language-, and structure-based approaches and informing model selection for protein engineering.