Nature Methods：同济大学史偈君团队开发三代测序检测RNA修饰的算法基准平台，为多种修饰检测提供权威指南

Core Viewpoint - The article discusses a systematic evaluation of computational tools for detecting multiple types of RNA modifications using Nanopore Direct RNA Sequencing (DRS), highlighting the performance differences and limitations of existing algorithms [1][2]. Group 1: Research Findings - A high-quality, single-base resolution benchmark dataset was constructed as a "gold standard" to evaluate 86 RNA modification detection algorithms based on DRS technology across four dimensions: accuracy, biological relevance, cross-sample generalization ability, and computational efficiency [2]. - The model retraining strategy significantly improved detection performance, with combined training on in vitro transcribed (IVT) RNA and real biological samples enhancing prediction accuracy and generalization, particularly for Ψ, m5C, and A-to-I modifications [4]. - m6A detection tools performed well overall, with models like Dorado and SingleMod excelling in qualitative and quantitative analysis, while most non-m6A modification detection tools struggled with quantitative accuracy and cross-sample generalization [5]. - Biological relevance is a crucial criterion, as ideal detection tools should not only have high accuracy but also align with known biological patterns; some tools showed discrepancies in predicted modification site distributions [5]. - Current tools face challenges in reliably distinguishing different modifications occurring on the same base, leading to "fuzzy predictions," which is a key area for future algorithm optimization [4]. - The retrained models can adapt to the iterative advancements in DRS technology, effectively addressing the challenges posed by changes in signal characteristics with the upgrade from RNA002 to RNA004 [6]. Group 2: Resource Development - To promote field development, the research team launched NaRMBench, an online resource platform that consolidates 12 key performance indicators of each tool into an interactive radar chart, aiding users in selecting and comparing analysis tools [8].