Skip to content

Quick Start

1. Prepare Parameters

Create or edit a params file:

tabnado-init

# tabnado parameters
# Required keys
target: MLLN
model_name: GANDALF
sweep_fraction: 0.2
gtf_file: data/gencode.vM25.annotation.gtf.gz
eval_chr: chr8
test_chr: chr9
output_dir: results

# Optional keys
dataset: data/dataset
# windows_bed: data/tss_windows.bed
n_sweeps: 10
logging: wandb
min_target: 1
min_features: 10
exclude_ips: ["AF4C", "MLLC"]
prefixes: ["CAT", "ChIP", "CM"]
window_size: 3000
step_size: 100
tile_size: 100
# chunk_size_rows: 1000000  # optional: limit rows per chunk for memory-constrained loading

2. Run Full Pipeline

tabnado-run --params params.yaml

3. Check Outputs

Look under:

results/<MODEL_NAME>_<TARGET>/

You should see:

  • best_hyperparameters.json
  • final_model/
  • figures/scatter_test.png
  • figures/embeddings_umap.png
  • shap/shap_mean_abs.csv
  • figures/shap_clustermap.png
  • shap/spatial_shap_by_offset_<target>.csv
  • figures/shap_spatial_heatmap_<target>.png
  • figures/shap_offset_line_<target>.png

4. Run on Slurm with a Container

Pull the image once on the cluster login node:

apptainer pull tabnado.sif docker://ghcr.io/cchahrour/tabnado:latest

If your cluster uses singularity instead of apptainer, use the same commands with singularity.

Note on quantnado:

  • If data_dir already contains cached parquet splits (dataset_train.parquet, dataset_eval.parquet, dataset_test.parquet), the pipeline can run from those cached files.
  • If those parquet files do not exist, tabnado-data needs quantnado available to build them from the raw dataset store.

Example single-job submission script (run_tabnado.sbatch):

#!/bin/bash
#SBATCH --job-name=tabnado
#SBATCH --partition=compute
#SBATCH --cpus-per-task=8
#SBATCH --mem=64G
#SBATCH --time=24:00:00
#SBATCH --output=logs/%x-%j.out

set -euo pipefail

cd /path/to/tabnado
DATA_DIR=/path/to/dataset
apptainer exec \
    --bind "$PWD:$PWD" \
    --bind "$DATA_DIR:$DATA_DIR" \
    --pwd "$PWD" \
    tabnado.sif \
    tabnado-run --params params.yaml

Submit with:

sbatch run_tabnado.sbatch

Example array job for multiple parameter files:

#!/bin/bash
#SBATCH --job-name=tabnado-array
#SBATCH --array=0-3
#SBATCH --cpus-per-task=8
#SBATCH --mem=64G
#SBATCH --time=24:00:00
#SBATCH --output=logs/%x-%A_%a.out

set -euo pipefail

cd /path/to/tabnado
DATA_DIR=/path/to/dataset
apptainer exec \
    --bind "$PWD:$PWD" \
    --bind "$DATA_DIR:$DATA_DIR" \
    --pwd "$PWD" \
    tabnado.sif \
    tabnado-run --params experiments/params_${SLURM_ARRAY_TASK_ID}.yaml