BEND 2.0 benchmark evaluates DNA language models
Original title: BEND 2.0: Benchmarking the Capabilities Needed to Design DNA
New benchmark tests genomic foundation models on eight tasks for predicting DNA edit effects and guiding sequence design.
Genomic foundation models are applied to propose DNA sequences and edits. Successful design requires capabilities not measured by current benchmarks. BEND 2.0 introduces eight tasks built on five experimental datasets to evaluate model predictions of DNA change effects.
- Tests prediction of DNA edit effects
- Eight tasks from five experimental datasets
- Linear readouts match or exceed native model outputs
- Functional post-training improves capabilities
- Model predictions reduce candidate sequences needed
From the original article
1 min read at huggingface.co
Genomic foundation models are increasingly used to propose DNA sequences and edits. But successful design requires capabilities that current benchmarks do not measure. Here we introduce BEND 2.
By Andreas Bjerregaard, Rachael DeVries, Dionysia Danai Brilli · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.