Benchmark RadarTadA-Bench
TadA-Bench is a fixed-data replay benchmark derived from 31 wet-lab rounds of TadA directed evolution. Models rank ~1M protein, DNA, or RNA sequence variants appearing only in later rounds, with scores as Spearman, Recall@10%, and nDCG@10%.
- Released
- 2026-05-29
- Readiness
- Runnable
- Primary field
- Health & Life Sciences
Why it matters
Standard random-split evaluation overestimates performance for iterative candidate prioritization. TadA-Bench provides a chronological replay protocol to measure whether models can transfer from earlier experimental rounds to future ones, a core requirement for agentic protein engineering.
Motivation
AI for scientific discovery is entering an agentic era, where protein-engineering systems are expected to prioritize future wet-lab experiments rather than merely fit static measurements.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.