smfnet-data-ont-mesc

Single-molecule footprinting with Oxford Nanopore sequencing (SMF-ONT) in DNMT triple-knockout (DNMT-TKO) mouse embryonic stem cells, the training data of the SMFNet model kundajelab/smfnet-ont-mesc. The files are those in the paper's Zenodo record (10.5281/zenodo.23407172, inputs/ont/); raw data are in GEO (GSE350193, GSE350195).

Data

Nuclei from four biological replicates were treated with the GpC methyltransferase M.CviPI and then the CpG methyltransferase M.SssI, which methylate cytosines in accessible DNA, and high-molecular-weight DNA was sequenced on PromethION R10.4.1 flow cells. Reads were basecalled and aligned to mm10 with Bonito v0.7.3 (dna_r10.4.1_e8.2_400bps_sup@v4.2.0, Minimap2 v2.24), and reads with a quality score below 7 were discarded. CpG and GpC methylation was called with the nf-smfont pipeline; calls with an absolute log-likelihood ratio below 2 are treated as missing, and reads with an overall methylation below 5% or above 95% were removed. See the paper's Methods for details.

File	Regions	Contents
ONT_mESC_acRegions.h5	77,030	active regions
ONT_mESC_bgRegions.h5	76,995	background regions

Format

Each file holds, for every region named chr:start-end (mm10, 1-based start, exclusive end, 2,114 bp):

• smf_mat/<region>: [reads, sites] integer calls: 1 unmethylated (protected, occupied), 0 methylated (accessible), -1 no call;
• smf_pos/<region>: the 0-based offsets of the columns' CpG and GpC cytosines within the region;
• read_id/<region>: the Nanopore read ID of each row.

SMFNet trains on 2,048-bp windows cropped from the regions and predicts the central 1,024 bp. The released model uses chr8 and chr16 for validation, chr1, chr3 and chr6 for testing, and the other chromosomes for training.

Usage

from huggingface_hub import hf_hub_download

paths = [
    hf_hub_download("kundajelab/smfnet-data-ont-mesc", name, repo_type="dataset")
    for name in ("ONT_mESC_acRegions.h5", "ONT_mESC_bgRegions.h5")
]

With SMFNet, pass them and the UCSC mm10 FASTA to its data configuration, e.g. data.cfg.h5_path=[...] and data.cfg.fasta_path=mm10.fa for scripts/train.py or scripts/sample.py.

Citation

Schwämmle, T., Diamant, N., Badayeva, Y., Kleinendorst, R., Kundaje, A. & Krebs, A. R. SMFnet reveals principles of combinatorial transcription factor assembly. bioRxiv (2026). https://doi.org/10.64898/2026.10.30.762954