RNAbpFlow training data and code

RNAbpFlow: Base pair-augmented SE(3)-flow matching for conditional RNA 3D structure generation This repository contains the code and resources for training, fine-tuning and inference of RNAbpFlow. Overview RNAbpFlow, a novel sequence- and base-pair-conditioned SE(3)-equivariant flow matching model for generating RNA 3D structural ensemble. Leveraging a nucleobase center representation, RNAbpFlow enables end-to-end generation of all-atom RNA structures without the explicit or implicit use of evolutionary information or homologous structural templates. RNAbpFlow is freely available at https://github.com/Bhattacharya-Lab/RNAbpFlow. Training paradigms In this repository, you will find the following training datasets and trained model checkpoints: Checkpoint Name Parameters Training Data Date Cutoff Fine-tuned RNA3DB.ckpt 16.9M Representative chains from RNA3DB-provided training split 2024-04-26 No CASP15.ckpt 16.9M Representative chains from entire RNA3DB set 2022-04-26 Yes CASP16.ckpt 16.9M Representative chains from entire RNA3DB set 2024-04-26 Yes RNAbpFlow: Training and Fine-Tuning Guide Download the RNAbpFlow-Train.tar.gz folder and extract the files Use mamba to create a virtual environment and install dependencies for RNAbpFlow conda install -n base -c conda-forge mamba mamba env create -f RNAbpFlow.yml Activate the virtual environment conda activate RNAbpFlow Make the main script executablechmod +x run.sh Train RNAbpFlow To train RNAbpFlow, run one of the following commands depending on the dataset or benchmark of interest. RNA3DB train-test split ./run.sh --dataset RNA3DB --gpu 0,1,2,3,4,5,6,7 Blind CASP15 experiment ./run.sh --dataset CASP15 --gpu 0,1,2,3,4,5,6,7 Blind CASP16 experiment ./run.sh --dataset CASP16 --gpu 0,1,2,3,4,5,6,7 Fine-Tune a trained Model To fine-tune RNAbpFlow using predicted base pairs, provide the path to an existing checkpoint. For CASP16 evaluation ./run.sh --dataset CASP16 --gpu 0,1,2,3,4,5,6,7 --finetune --ckpt /path/to/model.ckpt For CASP15 evaluation ./run.sh --dataset CASP15 --gpu 0,1,2,3,4,5,6,7 --finetune --ckpt /path/to/model.ckpt NOTE: All the pre-trained/fine-tunned checkpoints can be found in Checkpoints.tar.gz folder Inference using RNAbpFlow To generate sample 3D structures using RNAbpFlow Download and extract the Inference.zip folder and extract it. Detailed instructions for multi-GPU sampling are provided in README.md Alternatively, visit https://github.com/Bhattacharya-Lab/RNAbpFlow and follow the steps