99Science · AI & science

SCIENCE UNDER AUTOMATION

AI in Experimental molecular biology: capabilities and evidence

An agent can reason about a protocol while an experiment still depends on instruments, specimens and troubleshooting. Automation becomes more plausible where those physical steps are standardized and connected.

What has AI demonstrated?

40% lower protein-production cost reported

Test
GPT-5 / Ginkgo; six rounds, one sfGFP protein and one CFPS system.
Comparison
Best prior baseline cited by authors; no independent matched replication.
System and version
GPT-5 / Ginkgo cell-free protein synthesis · 2026-02-05
Evidence checked
2026-10-20

One protein (sfGFP), one CFPS system. Thousands of compositions are not thousands of independent protein workflows.

What does the field require?

Optimize a configured protein synthesis assay

Establish a new molecular mechanism

Research tasks

Interpret literature and plan a protocol

Information tasks have benchmark coverage; physical success requires a different test.

Execute and troubleshoot an experiment

Requires instrument access and evidence from the actual experimental setting.

Establish a causal mechanism

A convincing account depends on discriminating tests and independent validation.

Can AI replace the whole workflow?

The results above measure specific tasks. The field assessment also requires completion of the connected workflow, including the requirements listed here. A field-wide replacement percentage has not been established.

Read the assessment method

Sources and related evidence

Compare related fields

Computational genomics · Theoretical physics · Field ecology