RNovA End-to-End Workflow

This repository bundles the RNovA PathSearcher and SeqFiller inference modules plus post-processing (FDR and clustering) to run an open-PTM de novo workflow. The top-level entry point is RNovA.sh.

Contents

• RNovA_PathSearcher_Inference/: PathSearcher inference (stage 1, backbone path discovery). Developed by Zeping Mao (zeping.mao@uwaterloo.ca)
• RNovA_SeqFiller_Inference/: SeqFiller inference (stage 2, sequence completion). Developed by Zeping Mao (zeping.mao@uwaterloo.ca)
• workflow.py: clustering + alignment to derive top-k PTMs from PathSearcher results. Developed by Yonghan Yu
• FDR_stage1.py: node/path-level FDR filtering on PathSearcher outputs. Developed by Qianqiu Zhang
• FDR-stage2.py: sequence-level FDR filtering on SeqFiller outputs. Developed by Qianqiu Zhang
• decoy_spectrum_generator.py: decoy MGF generator (source: http://github.com/nh2tran/NovoBoard)

For module-specific details, see:

• RNovA_PathSearcher_Inference/README.md
• RNovA_SeqFiller_Inference/README.md

Dependencies

Module-specific dependencies are listed in each module's requirements.uv. In addition, the top-level workflow scripts require:

• Linux System with GPU
• Python 3.11+
• numpy, pandas, numba, tqdm, requests

Notes:

• workflow.py --use-unimod enables UniMod annotation and uses requests to fetch UniMod data.
• FDR-stage2.py imports RNovA_SeqFiller_Inference.utils.BasicClass, so SeqFiller must be installed/built (see its README for python setup.py build_ext --inplace).
• Checkpoint: https://zenodo.org/records/18352464

Clone With Submodules

This repository uses git submodules for the two inference modules. Clone with submodules so the folders are populated:

git clone --recurse-submodules https://github.com/zqq66/RNovA.git

If you already cloned without submodules:

git submodule update --init --recursive

Quick Start (End-to-End)

./RNovA.sh <in_dir> <useunimod:true|false> <k>

Arguments:

• in_dir: directory containing input .mgf files.
• useunimod: true to pass --use-unimod to workflow.py (enables UniMod annotation); false skips UniMod annotation.
• k: number of PTMs to keep in the top-k list.

Example:

./RNovA.sh ./data true 10

RNovA.sh runs:

1. decoy MGF generation
2. PathSearcher inference
3. FDR stage 1 filtering
4. clustering + alignment to get top-k PTMs
5. SeqFiller inference
6. FDR stage 2 filtering