ERA Discovers 40 Bioinformatics Methods, Beating Human Tools on Public Leaderboard
Updated
Updated · Nature.com · May 19
ERA Discovers 40 Bioinformatics Methods, Beating Human Tools on Public Leaderboard
6 articles · Updated · Nature.com · May 19
Nature reported that ERA—an AI system for writing scientific software—produced 40 new single-cell analysis methods that outperformed the best human-developed entries on a public benchmark.
The system targets a key research bottleneck: slow manual coding. It combines a large language model with tree search to iteratively improve a chosen quality metric and explore many candidate solutions.
ERA also generated 14 epidemiology models that beat the CDC ensemble and every other individual model in forecasting COVID-19 hospitalizations.
Beyond those headline results, it delivered expert-level software for geospatial analysis, zebrafish neural activity prediction, numerical integration and a new rule-based time-series forecasting approach.
The paper positions ERA as part of a broader push toward AI research agents that can turn scientific ideas into working, high-performing software faster.
Can AI scientists produce revolutionary breakthroughs or just faster incremental discoveries?
As AI automates research, will it level the scientific playing field or create a new class of 'AI-haves'?
Who legally owns scientific discoveries made by AI if they cannot be copyrighted?
Google Launches Gemini for Science and Publishes ERA AI in Nature: Transforming Scientific Discovery with Expert-Level AI (May 2026)
Overview
On May 19, 2026, Google marked a major milestone by publishing its Empirical Research Assistance (ERA) system in Nature and launching Gemini for Science for trusted testers. ERA is an advanced AI designed to enhance computational modeling, and it has demonstrated expert-level performance across many scientific fields. Built on ERA, the Computational Discovery engine helps researchers quickly test thousands of hypotheses, speeding up scientific progress. These developments reflect Google’s vision to make powerful AI tools widely accessible, empowering both new and experienced scientists to accelerate discoveries across diverse disciplines.