Appearance
Basenji2 Human Regulatory
Basenji2 predicts regulatory activity from DNA sequence using a convolutional TensorFlow model trained on human regulatory targets.
What it does
Liatir uses the official Calico Basenji2 human model files to predict regulatory signal from a DNA window. With a VCF file, it can compare reference and alternate windows for focused variant-effect exploration.
When to use it
Use Basenji2 when you want a managed local regulatory model that is lighter than very large long-context stacks and you are working with human regulatory sequence examples.
Inputs in Liatir
- Reference FASTA/FA/FNA file, or an inline DNA sequence.
- Optional VCF or VCF.GZ file.
- Output head:
Human. - Target index.
- Maximum variants to score.
Outputs
Liatir writes:
- regulatory signal CSV;
- BED signal track;
- optional variant score CSV and BED track;
- JSON summary;
- Results panels and provenance.
Hardware and installation
Basenji2 runs through a dedicated Python/TensorFlow runtime box. Liatir downloads the official human model weights, model parameters, and target table into that box.
CPU runs are possible for small tests. GPU acceleration is preferred for larger or repeated workflows.
Limits and cautions
The first Liatir integration exposes a practical target-index workflow. It does not yet include a friendly target-label browser, so target index 0 is the default starting point.
Treat signal tracks as model predictions. They are useful for comparing windows and variants, but they are not direct measurements from an experiment.