How ERA generates and optimises scientific software for computational experiments

ERA combines a Large Language Model (Gemini) with a tree search algorithm - Flat UCB Tree Search (FUTS), inspired by AlphaZero - to iteratively generate, execute, and score candidate programs. The input is a "scorable task": a problem description, a scoring metric, and training/validation data. ERA then searches scientific literature for research ideas, generates code implementing those ideas, and uses tree search to navigate the space of possible solutions, balancing exploration of new approaches against exploitation of promising ones.

The system does not produce a single answer. It generates a tree of software candidates, with each node representing a different algorithmic approach, and converges on expert-level solutions by iterating through thousands of code variants. The scoring metric drives the search; ERA keeps refining until it maximises the defined quality measure.

The benchmarks are concrete. In bioinformatics, ERA discovered 40 novel methods for single-cell RNA sequencing batch integration that outperformed the top human-developed methods on the scIB public leaderboard, where nearly 300 existing tools compete. In epidemiology, ERA generated 14 models that outperformed the CDC ensemble and all other individual models for forecasting COVID-19 hospitalisations at a state level across the United States. The system also demonstrated expert-level performance on satellite imagery analysis, neuroscience prediction, general time-series forecasting, and mathematical optimisation benchmarks.

ERA is open-source. The codebase, including the FUTS algorithm and benchmark notebooks, is available on GitHub (google-research/era). Eight published manuscripts apply ERA to specific scientific problems spanning epidemiology, climate, hydrology, economics, neuroscience, and combinatorics.

How BioSkepsis retrieves and reasons over biomedical literature for researchers

BioSkepsis retrieval is weighted by Gene Ontology terms, MeSH descriptors, gene symbols, and pathway relationships. A query about AMPK activation and its downstream effects on hepatic lipogenesis returns papers linked by the biological concepts involved, not just papers whose abstracts contain matching keywords. This biology-native retrieval layer is what separates BioSkepsis from both general-purpose LLMs and computational experimentation systems that treat biomedical papers as sources of algorithmic ideas rather than as primary evidence.

Answers are grounded in full text including methods, controls, and supplementary data. Every claim links back to the exact passage in the retrieved paper. When evidence is insufficient, BioSkepsis declines to answer rather than generating a plausible-sounding response. The research landscape graph classifies papers by structural role (Foundational, Hub, Bridge, and Novel) and draws on Semantic Scholar's 214M+ corpus for landscape expansion.

BioSkepsis also performs citation verification: a seven-step pipeline (Steps A through G) that audits whether cited papers in an existing document actually support the claims they are attached to. This is a forensic capability; it checks the integrity of literature use, not just its availability.

The EFEVRE ecosystem: physical execution, documentation, and interpretation in one loop

BioSkepsis is one of three products built by EFEVRE TECH LTD. The other two address problems that ERA does not attempt to solve.

AMGEL (AutoMated GEneral Laboratory) is a patent-pending robotic platform that automates all mainstream wet-bench procedures on one device: pipetting, centrifugation, heating/cooling, magnetic bead isolation, cold storage, and PCR. It runs 860+ pre-set protocols with 24/7 unattended operation. AMGEL physically executes experiments; ERA automates only the computational side.

VITALE (Versatile Integrated Technology Advancing Life-Science Exploration) is the software layer that records every step, parameter, timestamp, and result during AMGEL operation. Its Protocol Designer, Personal Labbook, Scheduling Calendar, and Protocol Library with community scoring ensure that no experimental detail goes undocumented. VITALE addresses the documentation pillar of the reproducibility crisis: even when experiments are performed correctly, poor reporting in publications makes replication structurally impossible.

Together, the three products form a closed loop: AMGEL removes human error from execution, VITALE removes human error from documentation, and BioSkepsis removes human bias from interpretation. ERA addresses a fourth dimension - computational experimentation - but does not touch the physical, documentation, or literature-interpretation layers.

Feature comparison: BioSkepsis (EFEVRE ecosystem) vs ERA (Google DeepMind)

Where ERA leads: automated computational experimentation across scientific disciplines

ERA's primary advantage is in automating the most labour-intensive phase of computational science: writing, testing, and refining the software that implements a scientific idea. The tree search architecture is the key differentiator. Rather than generating a single piece of code and hoping it works, ERA explores a branching tree of algorithmic variants, each scored against the defined quality metric, and converges on solutions that outperform expert-written alternatives.

The single-cell RNA-seq results are the strongest demonstration of this capability. Batch integration in scRNA-seq is a well-studied problem with nearly 300 existing human-developed tools and multiple public benchmarks. ERA discovered 40 novel methods that outperformed every one of those tools on the scIB leaderboard. This is not incremental improvement; it represents a qualitative expansion of the solution space that human developers had not explored.

The epidemiological forecasting results carry immediate public health significance. ERA generated 14 models that outperformed the CDC ensemble and all other individual models for forecasting COVID-19 hospitalisations at a state level. Google has since joined the CDC's live forecasting challenges for flu, COVID-19, and RSV, submitting weekly ERA-generated predictions for every U.S. state.

ERA's position within the Gemini for Science ecosystem amplifies its utility. Computational Discovery, the prototype built with ERA and AlphaEvolve, is available through Google Labs and allows researchers to define a scientific problem, provide a scoring metric, and let the system explore thousands of code variations in parallel. For computational scientists whose bottleneck is the iteration cycle of writing, testing, and refining analysis code, this is a direct productivity accelerator.

Where BioSkepsis leads: citation integrity, reproducibility, and evidence-grounded reasoning in the life sciences

BioSkepsis occupies territory ERA does not enter. The two systems operate on different substrates entirely: ERA operates on code and data; BioSkepsis operates on scientific text and evidence chains.

Citation verification, the seven-step audit pipeline that checks whether cited papers actually support the claims they are attached to, has no equivalent in the ERA system. ERA reads literature to extract algorithmic ideas for code generation; it does not audit whether papers cited in existing publications correctly support their associated claims. For researchers, reviewers, and editors concerned with the accuracy of the scientific record, this is a meaningful gap that ERA was never designed to fill.

The research landscape graph, with its classification of papers into Foundational, Hub, Bridge, and Novel roles via citation network analysis, provides structural understanding of a field that ERA's code-generation pipeline does not surface. A researcher mapping the state of a sub-discipline, identifying which seminal papers anchor the field and which bridge papers connect separate domains, needs this kind of analysis.

Biology-native retrieval is the most architecturally consequential difference. BioSkepsis indexes papers by Gene Ontology terms, MeSH descriptors, gene symbols, and pathway relationships. A query about AMPK activation via LKB1 returns different papers than a query about AMPK activation via CaMKK2, because the retrieval layer understands the biological distinction. ERA's literature search is oriented toward finding methods and algorithmic approaches, not biological mechanisms.

The EFEVRE ecosystem's physical layer addresses problems no computational system can. AMGEL runs 860+ laboratory protocols autonomously, 24 hours a day. VITALE records every parameter. ERA automates computation; the EFEVRE ecosystem automates execution, documentation, and interpretation.

Where BioSkepsis and ERA overlap, and how they complement each other in biomedical research

The overlap is narrow. Both systems use scientific literature as input, but they use it for entirely different purposes. ERA reads papers to extract algorithmic ideas that inform code generation. BioSkepsis reads papers to produce citation-grounded evidence syntheses, map field structure, and verify citation integrity. A researcher querying ERA with a bioinformatics problem gets optimised code. A researcher querying BioSkepsis with a biological question gets an auditable evidence synthesis with PMIDs.

The complementary workflow is clearer here than in most AI-for-science comparisons, because the tools are so architecturally distinct. A computational biologist working on single-cell genomics could use BioSkepsis to survey the literature on batch-effect correction methods, understand which approaches are supported by robust experimental evidence, identify knowledge gaps, and verify the citations in their manuscript draft. They could then use ERA (or the Computational Discovery prototype) to generate and optimise novel batch-correction code that outperforms existing tools. The literature reasoning happens in BioSkepsis; the code optimisation happens in ERA; neither duplicates the other's core job.

The EFEVRE ecosystem extends this further. If the single-cell data comes from a physical experiment, AMGEL standardises the upstream wet-lab steps (cell isolation, RNA extraction) and VITALE documents every parameter. The result is a fully traceable pipeline: reproducible physical execution (AMGEL), automated documentation (VITALE), optimised computation (ERA), and verified literature grounding (BioSkepsis).

Who should use which tool in biomedical and life-science research

Frequently asked questions