CodLncEnPred

Table of Contents

• Online Web Server
• Project Goal
• Datasets
• Environment Deployment (offline)
• How to run the pipeline
• Output layout
• Notes & tips

Online Web Server

If you prefer not to run the pipeline locally, try the online prediction platform:

🌐 Access: EnsembleLnc Web Server — upload or paste FASTA sequences and get ensemble predictions in one click.

Project Goal

This project provides a robust ensemble inference pipeline for identifying and classifying coding long non-coding RNAs (codlncRNAs). The pipeline runs multiple base classifiers, merges their outputs, and applies a meta model to produce final prediction scores.

Integrated base models / tools (examples): CPAT, ABLNCPP, LncReader, mRNN, RNAsamba, IRSOM, DNABERT-2, RiNALMo, RNA-FM, RNABERT.

Datasets

All datasets used for model training, validation, and cross-species testing are available on our download page:

📥 Download: EnsembleLnc Datasets Download

Key collections:

• Four-level Confidence codlncRNA Dataset — sequences labelled at four confidence levels.
• Cross-species codlncRNA Dataset — sequences from multiple species for testing generalization.

Environment Deployment (offline)

To avoid dependency conflicts, we provide packaged Conda runtime environments as .tar.gz archives. Extract them locally — no internet required during extraction.

Provided environment archives (examples):

dna.tar.gz
irsom.tar.gz
lncrna.tar.gz
mRNN.tar.gz
rinalmo.tar.gz
rnafm.tar.gz
rnasamba.tar.gz

Deployment steps

1. Create the project conda_envs directory (the pipeline looks here by default):

mkdir -p conda_envs

2. For each archive, create the target folder and extract. Example for lncrna:

mkdir -p conda_envs/lncrna
tar -xzf env_packs/lncrna.tar.gz -C conda_envs/lncrna

3. Repeat for all provided archives. After extraction, each tool should be callable from conda_envs/<envname>/bin.

By default, the pipeline script searches ${PROJECT_ROOT}/conda_envs/<env>/bin. If your Conda environments are located elsewhere, set CONDA_ENVS_DIR before running, for example:

export CONDA_ENVS_DIR=/home/you/miniconda3/envs

You may also override a single environment path:

export ENV_RINALMO_BIN=/custom/path/rinalmo/bin

How to run the pipeline

run_pipeline.sh automates base model inference, result merging, and meta model prediction.

1. Prepare input

Provide a FASTA file, e.g. test.fa.

2. Change to the project base (where run_pipeline.sh lives)

cd /path/to/base_model

3. Run the pipeline

Usage

./run_pipeline.sh <test_fa> [out_root]

• <test_fa> — required: path to input FASTA (e.g., ./test.fa)
• [out_root] — optional: output root directory (defaults to current directory)

Example

./run_pipeline.sh ./test_20sequences.fa ./results

Output layout

After a successful run, results are saved under a timestamped directory:

<out_root>/pipeline_results/YYYYmmdd_HHMMSS/
  ├── input.fa                # copy of the original input
  ├── logs/                   # per-tool execution logs
  ├── predict/                # intermediate predictions from each base model
  └── final_pred.tsv          # final ensemble predictions (scores & labels)

Open final_pred.tsv to inspect final classification scores and labels.

Notes & tips

• The pipeline does not require conda activate if you provide the environment bin/ directories — it calls the interpreters and executables directly.
• For GPU-based models, ensure the environment contains GPU-compatible packages (PyTorch/TensorFlow) that match your system's CUDA drivers.

License

This source code is licensed under the MIT license.