Impact of training data on deep learning-based RNA 3D structure quality assessment [Datasets]

Impact of Training Data on Deep Learning-Based RNA 3D Structure Quality Assessment - Datasets This repository contains the datasets and pre-trained models accompanying the publication of the same title. The goal of the research was to validate and cross-train available architectures for quality assessment (QA) of RNA 3D structures. Archive Overview 1. ares/lociparse/rna3dqa Charts presenting dataset details and analysis. Note: some charts were not used in the final publication. 2. data 2.1 fasta_cd_hit_results - Redundancy check results obtained using CD-HIT. 2.2 lddt_training - Prepared ARES/lociPARSE/RNA3DQA datasets for per-nucleotide lDDT training. 2.3 rmsd_training - Prepared ARES/lociPARSE/RNA3DQA datasets for training using the RMSD metric. 2.4 rnapuzzle_analysis - Results of the model standardization process and final RMSD calculations. 2.5 rnapuzzle_benchmark - Full benchmark dataset, including a CSV file listing the selected solutions used for lDDT and RMSD calculation. 2.6 rnapuzzle_full - All available RNA-Puzzles contest data, along with the RMSD calculation configurations investigated. Here is located also structures_repaired directory, where are structures cured with rna-tools. 3. models 3.1 ARES - All trained models, including a Docker image archived as .tar. 3.2 lociPARSE - All trained models. 3.3 RNA3DCNN - All trained models. 4. rnapuzzle-gemini Scraped results from the RNA-Puzzles website, used to validate the plausibility of our results. 5. test All pipelines used to generate performance comparisons and dataset analysis charts. ARES/lociPARSE/RNA3DQA/RNA3DCNN contain model results.