RNAInvBench: Benchmark for the RNA Inverse Design Problem

Published: 17 Jun 2024, Last Modified: 22 Jul 2024ICML2024-AI4Science PosterEveryoneRevisionsBibTeXCC BY 4.0
Keywords: RNA, RNA Inverse Design, RNA Design, Benchmark, Inverse Design
TL;DR: Open-source benchmark for the RNA inverse design problem.
Abstract: The RNA inverse folding problem involves a challenging one-to-many problem space, where many RNA sequences can fold into a provided secondary structure. The primary methodology for generating a varied set of solutions involves an algorithmic approach to navigate the search space. Optimising RNA sequences for specific functions and enhancing their efficacy has a broad array of applications across medicine, biotechnology and synthetic biology. Despite a range of impressive algorithms, the lack of a unified platform for comparing algorithms and techniques hinders comprehensive analysis. To address this, we propose RNAInvBench, an open-source framework for Inverse RNA Design, which comprises of two focused tasks, pseudoknot-free and pseudoknot-inclusive design, two essential tasks for understanding and benchmarking model efficacy and accuracy. We further collate and organise key datasets for future model training and testing, and provide several detailed metrics for evaluating the success, function and efficiency of the generated sequences. We provide package environments through dockerfiles, to facilitate easy access for researchers. All source code is publicly available for use and download at: https://anonymous.4open.science/r/RNADesignPipeline-0572/
Submission Number: 243
Loading