TadA-Bench: A Million-Variant Benchmark for Future-Round Discovery Toward Agentic Protein Engineering

2026-05-29

AI Summary

Current AI models, predominantly trained on static datasets, struggle to effectively prioritize wet-lab experiments in protein engineering. This work introduces TadA-Bench, a temporal benchmark comprising over one million variants from 31 rounds of TadA directed evolution, which uniquely preserves experimental chronology through a fixed-data replay task that emphasizes cross-round generalization. Evolutionary coverage is proposed as a key metric of informative diversity. The authors develop a Seq2Graph graph neural network to jointly model noisy enrichment measurements across multiple rounds, enabling alignment of DNA, RNA, and protein views. While models perform well under random data splits, their performance markedly degrades in forward-looking tasks—specifically, ranking future-round variants and selecting candidates under limited experimental budgets—highlighting the benchmark’s rigor and relevance for surrogate-driven protein discovery.

Abstract

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. We introduce TadA-Bench, a million-variant wet-lab replay benchmark from 31 TadA directed-evolution rounds for future-round discovery toward agentic protein engineering. TadA-Bench preserves the campaign chronology and defines a fixed-data replay task: given earlier experimental rounds, models rank variants that appear only in later rounds. It provides aligned DNA, RNA, and protein views, and uses Seq2Graph, a graph-based label-unification pipeline, to reconcile noisy enrichment measurements into consistent cross-round activity labels. Random-split controls show strong interpolation, but future-round ranking and finite-budget candidate selection are much weaker. Controlled analyses suggest that evolutionary coverage is more informative than local data density, positioning TadA-Bench as a reproducible wet-lab replay substrate for future-round discovery toward agentic protein engineering; the data and code are released on Hugging Face and GitHub.

Problem

• agentic protein engineering
• future-round discovery
• wet-lab replay benchmark
• TadA directed evolution
• protein variant ranking

Innovation

• agentic protein engineering
• future-round discovery
• wet-lab replay benchmark
• Seq2Graph
• evolutionary coverage