RawMod

Pretrained models for RawMod: Modification-Agnostic Detection of DNA Base Modifications from Nanopore Signal Pileups.

RawMod detects modified DNA bases directly from raw nanopore signals. One model detects 5mC, 5hmC, 6mA, 4mC and 5hmU, and it needs no unmodified control.

Code, installation, usage and full documentation: https://github.com/STORMgroup/RawMod

Files

Path	Use
rawmod.pt	The released model. Use this one to score new data.
region_holdout/	Models trained without one region of every genome. Used for the accuracy numbers of the paper. loco_<mod>.pt also withholds one modification type. seeds/ holds two more training seeds.
leaveout/	Models trained without one organism or library. Used for the paper's supplementary leave-out benchmark.

All files share one architecture (230,697 parameters). They are PyTorch state dictionaries for the code in the GitHub repository. SHA256SUMS lists the checksums of every file.

Download

hf download STORMResearch/RawMod --local-dir checkpoints

Citation

The RawMod paper is under review. Until it is published, please cite the GitHub repository: https://github.com/STORMgroup/RawMod.