← Datasets & benchmarks

LLM-bioseq-probe

LLM-bioseq-probe is a datasets & benchmarks tool whose code is published by XinmingTu on GitHub, first public in 2026. API-based probe for protein and DNA sequences in LLM training data: next-amino-acid accuracy on 668 little-studied DUF-family proteins versus residue-shuffled controls, with test sets, all replies and a CLI.

Visit website ↗

CategoryDatasets & benchmarks
Developer (GitHub)XinmingTu
Websitexinmingtu.cn/blog/2026/claude-opus-5-5-protein-sequences
GitHubXinmingTu/LLM-bioseq-probe
Released2026-11-17 (source)
Code licenseMIT
Weights licenseno model weights involved
What this allowsPermissive (commercial OK)
Use, modify and ship it inside closed-source commercial products; keep the license text and copyright notices. What do these mean?
License sourceraw.githubusercontent.com/XinmingTu/LLM-bioseq-probe/main/LICENSE · checked 2026-11-18
PricingOpen source
Tags#benchmark#protein-language-models#llm-evaluation#training-data#dna
Last checked · links verified

Can I use LLM-bioseq-probe commercially?

Yes. The recorded license is permissive, so closed-source commercial use is allowed if you keep the notices.

What the license types mean · not legal advice.

Specs

Size668 proteins from 167 DUF Pfam families (5 cuts each, 6,680 questions with shuffled controls); 257 families in the extended family-size set
Modality—
AccessGitHub repo (TSV test sets, gzipped JSONL replies); probe-model CLI for Anthropic, OpenAI, Gemini, Fireworks APIs
GitHub stars9
Interface—

Compare with…

Pick 1–3 other Datasets & benchmarks tools to compare side by side with LLM-bioseq-probe.

Select 1–3 tools

Related

OMol25

Meta FAIR's open dataset of DFT calculations on diverse molecules, with evaluations and baseline models.

Datasets & benchmarksOpen sourceUnknown#dft

ML-ready datasets and benchmarks for therapeutics.

Datasets & benchmarksOpen sourcePermissive (commercial OK)#benchmark

ProteinGym

Benchmarks for protein fitness and mutation effect prediction.

Datasets & benchmarksOpen sourcePermissive (commercial OK)#benchmark