RNAJog：Fast Multi-objective RNA Optimization with Autoregressive Reinforcement Learning

RNAJog is a tool designed for optimizing the coding sequence (CDS) of mRNA to achieve high protein expression levels. RNAJog can generate codon sequences with high codon adaptation index (CAI) and low minimum free energy (MFE), ensuring enhanced translational efficiency and mRNA stability. This tool enables users to optimize mRNA sequences for their target proteins or existing mRNA sequences.

Access RNAJog online: RNAJog Web Application

Prerequisites

Before installing RNAJog, ensure that you have Conda installed. You can download either:

• Miniconda
• Anaconda

Installation

To install RNAJog, follow these steps:

git clone https://github.com/kxstd/RNAJog.git
cd RNAJog
conda env create -f environment.yml
conda activate rnajog

We recommend you to run RNAJog on Linux.

Download Model Parameters

Download the required model parameters from this link and extract them into the RNAJog project directory.

Usage

Command-Line Arguments

RNAJog provides multiple options to customize the optimization process. Below are the available arguments:

Argument	Type	Default	Description
--device	str	cuda	Device to use (cpu or cuda).
--seed	int	0	Random seed for reproducibility.
--model	str	RNAJog	Optimization model (RNAJog or RNAJog_zero).
--input_type	str	rna	Input type (rna or protein).
--data_path	str	data/test/rna.txt	Path to input data file.
--codon_usage_freq_table_path	str	./codon_usage_freq_table_human.csv	Path to the codon usage frequency table.
--mfe_weight	float	0.15	Weight of MFE in optimization (MFE-CAI balance).
--sample_method	str	sample	Sampling method (greedy or sample).
--sample_size	int	1	Number of generated samples.
--sample_temperature	float	0.01	Temperature parameter for sampling.
--save_path	str	result/	Directory to save output.
--ban_seqs	str	""	Forbidden subsequences in output.

Running RNA Optimization

To optimize an RNA sequence, use:

python run.py --input_type rna --data_path data/test/rna.txt --mfe_weight 0.15 --device cuda --model RNAJog --ban_seqs "CUCGAG;GCUCUUC"

For protein sequence optimization:

python run.py --input_type protein --data_path data/test/protein.txt --mfe_weight 0.15 --device cuda --model RNAJog

Output

The optimized RNA sequences are saved in a CSV file located in the specified save_path. The output file contains the following columns:

• id: Sample identifier
• sample_method: Sampling method used
• length: Sequence length
• mfe_cai_weight: MFE-CAI weight used in optimization
• mfe: Minimum Free Energy of the sequence
• cai: Codon Adaptation Index
• seq: Optimized RNA sequence

Example Output (output.csv):

id,sample_method,length,mfe_cai_weight,mfe,cai,seq
0,input,300,-1,-125,0.89,UAUGCGUAGC...
1,sample,300,0.15,-142,0.91,UAUGCGUAGC...

Citation

If you find the tool useful in your research, please cite our paper:

Jiaqi Huang et al. Fast Multi-objective RNA Optimization with Autoregressive Reinforcement Learning. In submisson.