Malinois

Malinois predicts cis-regulatory activity of 200 bp human sequences in K562, HepG2 and SK-N-SH.

Machine-guided design of cell-type-targeting cis-regulatory elements (Gosai et al., Nature 2024).

This repository holds the published checkpoint (20211113_021200), converted to safetensors from gs://tewhey-public-data/CODA_resources/ with no retraining or modification.

For variant effect prediction, use MPAC instead.

Usage

from modeling_malinois import MalinoisModel

model = MalinoisModel.from_pretrained("saarantras1/malinois").eval()

preds = model.predict(["ACGT" * 50])   # (n, 3): K562, HepG2, SKNSH

Use predict rather than calling the model directly: it adds the MPRA vector context the model was trained with (a bare 200mer is not valid input) and averages over both strands. Skipping either step returns plausible-looking but wrong numbers instead of an error.

Note on strands: predict reverse-complements the 200 bp insert and re-flanks it in the forward orientation, following src/vcf_predict.py in the upstream code base. The CODA tutorial notebook instead flips the assembled 600 bp construct. Both appear in upstream code; this repository uses the former.

Citation

@article{gosai2024coda,
  title   = {Machine-guided design of cell-type-targeting cis-regulatory elements},
  author  = {Gosai, Sager J. and Castro, Rodrigo I. and Fuentes, Natalia and
             Butts, John C. and Mouri, Kousuke and Alasoadura, Michael and
             Kales, Susan and Nguyen, Thanh Thanh L. and Noche, Ramil R. and
             Rao, Arya S. and Joy, Mary T. and Sabeti, Pardis C. and
             Reilly, Steven K. and Tewhey, Ryan},
  journal = {Nature},
  year    = {2024},
  doi     = {10.1038/s41586-024-08070-z}
}

License

MIT