modScore is a Python-based tool for filtering RNA modification calls from Oxford Nanopore Technologies (ONT) Dorado and modkit.
It combines deep learning and known modification sites to reduce false positives.

• m⁶A
• m⁵C
• Ψ (Pseudouridine)
• Inosine

• hg38
• mm10

Clone the repository:

git clone https://github.com/uedaLabR/modScore.git
cd modScore

Download source files:

wget https://zenodo.org/records/15369045/files/source.zip/content .
unzip source.zip

Install dependencies (Python 3.10 or later recommended):

pip install --no-cache-dir \
    numpy==1.24.4 \
    tensorflow==2.15 \
    numba==0.60.0 \
    pandas==2.2.3 \
    pysam==0.22.0 \
    click==8.0.4 \
    scikit-learn==1.5.2

Run the program:

python MSCmd.py <command> [OPTIONS]

Pull the prebuilt image:

docker pull karkinos/modscore_v01:latest

python MSCmd.py filter \
  --bed input.bed \
  --bed_out filtered_output.bed \
  --source_path path/to/known_sites \
  --genome hg38

Arguments:

• --bed: Input BED file (from Dorado/modkit)
• --bed_out: Output filtered BED file
• --source_path: Directory containing known modification site BEDs
• --genome: Genome version (default: hg38)

➡️ Outputs both a filtered BED and a *_stats.txt summary.

python MSCmd.py reflectToBam \
  --bamin input.bam \
  --bamout output.bam \
  --filter_bed filtered_output.bed
  --genome_ref genome hg38.fa

Arguments:

• --bamin: Original BAM file
• --bamout: Output BAM with updated ML tags
• --filter_bed: BED file from the filter command
• --genome_ref: path to reference genome fasta file (fai file required)

python MSCmd.py trainSequenceClassification \
  --source_path training_data_dir \
  --genome hg38 \
  --fp_ivtpath ivt_data_dir \
  --outhistory train_log.txt \
  --weightpath model_weights.h5

Arguments:

• --source_path: Directory of training data
• --fp_ivtpath: Directory of positive (IVT) data
• --outhistory: Training history log output
• --weightpath: Output path for model weights

• Python 3.10
• Ubuntu 22.04
• TensorFlow 2.15 (GPU-enabled)
• ONT Dorado 0.9.1 + modkit 0.4.5

• Docker Hub: karkinos/modscore_v01