Appearance
Enformer Regulatory Prediction
Enformer is a long-context DNA model for predicting regulatory activity from genomic sequence.
What it does
Liatir uses Enformer to read a large DNA window and produce regulatory signal tracks. When a VCF file is provided, Liatir can compare the reference and alternate windows to estimate how much a variant changes the predicted signal.
When to use it
Use Enformer when you want to explore regulatory effects from DNA sequence and you need longer context than a short embedding model provides.
Inputs in Liatir
- Reference FASTA/FA/FNA file, or an inline DNA sequence.
- Optional VCF or VCF.GZ file for variant scoring.
- Output head, usually
Human. - Target index. Start with
0for a first test, then choose a more specific track once target labels are exposed in the UI.
Outputs
Liatir writes:
- predicted regulatory signal as CSV;
- BED genome track for the signal;
- optional variant score CSV and BED track;
- JSON summary;
- Results panels and provenance.
Hardware and installation
Enformer is TensorFlow-based and uses a large input window. CPU execution can be slow. A GPU-capable TensorFlow backend is preferred for repeated scoring.
Liatir installs the runtime as a separate AI Model box, including TensorFlow, TensorFlow Hub, and the Enformer TFHub asset cache.
Limits and cautions
Predictions are useful for exploration and prioritization, not automatic biological conclusions. Always check the input reference, selected target index, model notes, and provenance.
The current targetIndex is technical. Start with 0 for a smoke test, but do not make biological claims until you know which target the selected index represents.