ai4nucleome-glmap 0.0.1

GLMap: Profiling genomic language models as individuals in a population (placeholder release; full v1.0.0 forthcoming with the paper)

GLMap

Placeholder release (v0.0.1). The full v1.0.0 will be released together with the accompanying paper.

GLMap (Genomic Language Model Map) is a training-free, architecture-agnostic framework for representing and comparing genomic language models (GLMs) by their likelihood responses over a fixed panel of DNA sequences. Applied to 123 publicly available GLMs scored on a panel of 10,000 DNA probes, GLMap places autoregressive and masked-language models in a common space, yields model distances stable to the choice of probes, and predicts downstream task performance with Spearman ρ ≈ 0.7.

Status

This 0.0.1 release reserves the glmap name on PyPI. The full library — including model loaders, scoring code, panel construction, and the clip + double-center matrix pipeline — is in active preparation:

• Paper submitted
• Code repository public release
• PyPI v1.0.0

Track progress at https://github.com/ai4nucleome/GLMap.

Citation

If you wish to cite GLMap before the v1.0.0 release, please cite the preprint:

Hou, Y., Long, W., Su, H., Feng, J., Zhang, Y.
Profiling genomic language models as individuals in a population.
(In submission, 2026.)

License

Apache-2.0.