<strong>PlantC2U:</strong> <strong>Convolutional neural network based deep learning of cross-species sequence landscapes predicts plastid C-to-U RNA editing in plants </strong>

Table S1. The detailed information of plastid RNA editing sites collected in REDIdb 3.0.Table S2. The summary of the plastid C-to-U RNA editing sites dataset.Table S3. The performance of PlantC2U in independent test based on the different flanking sequence and the ratio of positive and negative instances from primary negative samples.Table S4. The performance of PlantC2U in independent test based on the different flanking sequence and the ratio of positive and negative instances from sequence clusters.Table S5. Prediction of C-to-U RNA editing sites based on the same testing data using CNN, RF, SVM, and PREPACT tools.Table S6. The C-to-U RNA editing sites identified by RNA-seq datasets were further predicted by PREPACT and PlantC2U tools.Table S7. The C-to-U RNA editing sites identified by RNA-seq and WGS data were further predicted by PlantC2U tool.Table S8. Identification of plastid C-to-U editing sites across eight tissues in <i>K. obovata</i>.Table S9. Annotation and functional prediction of plastid C-to-U editing sites on the protein-coding genes in <i>K. obovata</i>.Table S10. Identification of plastid C-to-U editing sites in <i>K. obovata</i> leaf tissues under chilling events.Table S11. The functional enrichment analysis of four DEGs (<i>psbA</i>, <i>psbM</i>, <i>rps12 </i>and <i>ndhA</i>)Table S12. The primers used for PCR assays.Table S13. The detailed information of 50 RNA-seq datasets, covering 4 tissues from 24 mangrove species.Table S14. Identification of plastid C-to-U editing sites in other 24 mangrove species.Figure S1. The evaluation of model performance under different combinations of negative and positive samples.Figure S2. Sequences clustering and nucleotide context flanking the C-to-U editing sites.Figure S3. <i>In silico</i> mutagenesis analysis and SNV identified from 24 <i>K. obovata</i> RNA-seq datasets.Figure S4. Quality control of the cleaned reads and the RNA editing level of plastid genes.Figure S5. SNV identified in 50 mangrove transcriptomes.Figure S6. The experimental verification of 20 C-to-U RNA editing events in <i>K. obovata</i>. Red boxes indicate the RNA editing sites in the cDNA.Figure 1. A schematic diagram of prediction of plastid C-to-U RNA editing sites. The 100-nt flanking sequences surrounding the target cytidines are extracted and converted into a 5×100 binary matrix as the input for PlantC2U. The CNN architecture includes 2 one-dimensional convolution layers (Conv1D), a max pooling layer, and a flatten layer. The nonlinear softmax activation function is used to calculate the prediction probability of C-to-U RNA editing. Figure 2. The evaluation of model performance under different combinations of negative and positive samples. (A) The model performance based on the Matthew's Correlation Coefficient (MCC) using RNS1 and RNS2 as negative samples, respectively. RNS1 represents random negative sequences from primary negative samples. RNS2 represents random negative sequences from different sequences clusters. The Ratio is positive-negative samples ratio. (B) The receiver operating characteristic (ROC) curves show the better performance of convolutional neural network (CNN) than the random forest (RF), support vector machine (SVM) models and PREPACT tool. (C) Evaluation of model performance based on the area under the precision-recall curves (AUC-PRC). Figure 3. Representative map of the plastid genome of mangrove plant <i>K. obovata</i>, including 83 genes, 8 rRNAs, 37 tRNAs, and 35% GC content. IRA and IRB indicate inverted repeat regions. LSC is large single copy. SSC is small single copy. Figure 4. Identification of plastid RNA editing sites in K. obovata. (A) Overall C-to-U RNA editing level in different <i>K. obovata</i> tissues. Values indicate mean ±SD of three biological replicates. The significant difference of RNA editing level were represented by different lowercases through one-way ANOVA (P &lt; 0.05). (B) SNV identified from K. obovata leaf transcriptomes. The bar plots show the percentage of each SNV type (e.g., C&gt;T, CT) identified in each K. obovata transcriptome. (C) The box plots show the overall C-to-U RNA editing level in cold tolerant (CT) and non-cold tolerant (NCT) transcriptomes. (D) The box plots of the expression level of four genes (psbA, psbM, rps12 and ndhA) in CT and NCT transcriptomes. * represent 0.01 &lt; P ≤ 0.05, ** represent 0.001 &lt; P ≤ 0.01. (E) The functional enrichment analysis of four genes (psbA, psbM, rps12 and ndhA) using STRING website tool. (F) The C-to-U RNA editing level of C371 and C991 sites from ndhA in CT and NCT transcriptomes. Figure 5. A workflow for identification and annotation of plastid RNA editing sites. (A) A pipeline includes quality control of raw sequencing data, clean read mapping, RNA editing sites calling, and C-to-U RNA editing sites annotation. (B) Number of C-to-U RNA editing sites identified from different tissues of 24 mangrove species. Figure 6. Interface of the online predictor, PlantC2U, on plants plastid C-to-U RNA editing sites.

表S1. 收录于REDIdb 3.0的质体RNA编辑位点详细信息
表S2. 质体C-to-U RNA编辑位点数据集汇总
表S3. 基于不同侧翼序列及来自原始阴性样本的正负样本比例的独立测试中PlantC2U的表现
表S4. 基于不同侧翼序列及来自序列聚类集的正负样本比例的独立测试中PlantC2U的表现
表S5. 使用卷积神经网络（Convolutional Neural Network, CNN）、随机森林（Random Forest, RF）、支持向量机（Support Vector Machine, SVM）及PREPACT工具，基于同一测试集预测C-to-U RNA编辑位点的结果
表S6. 经RNA-seq数据集鉴定的C-to-U RNA编辑位点，经PREPACT与PlantC2U工具进一步预测的结果
表S7. 经RNA-seq与全基因组测序（Whole Genome Sequencing, WGS）数据鉴定的C-to-U RNA编辑位点，经PlantC2U工具进一步预测的结果
表S8. <i>K. obovata</i> 8种组织中的质体C-to-U编辑位点鉴定结果
表S9. <i>K. obovata</i> 蛋白编码基因上质体C-to-U编辑位点的注释与功能预测
表S10. 低温胁迫下<i>K. obovata</i>叶片组织中的质体C-to-U编辑位点鉴定结果
表S11. 4个差异表达基因（<i>psbA</i>、<i>psbM</i>、<i>rps12</i>与<i>ndhA</i>）的功能富集分析结果
表S12. PCR实验所用引物信息
表S13. 50套RNA-seq数据集的详细信息，涵盖24种红树植物的4种组织
表S14. 其余24种红树植物的质体C-to-U编辑位点鉴定结果
补充图S1. 正负样本不同组合下的模型性能评估
补充图S2. C-to-U编辑位点侧翼序列的聚类与核苷酸上下文特征
补充图S3. <i>In silico</i> 计算机模拟诱变分析及从24套<i>K. obovata</i> RNA-seq数据中鉴定的单核苷酸变异（Single Nucleotide Variant, SNV）
补充图S4. 清理后读段的质量控制及质体基因的RNA编辑水平
补充图S5. 50个红树植物转录组中的单核苷酸变异鉴定结果
补充图S6. <i>K. obovata</i>中20个C-to-U RNA编辑事件的实验验证结果。红色方框标注cDNA中的RNA编辑位点
图1. 质体C-to-U RNA编辑位点预测流程示意图。提取目标胞嘧啶周围100 nt的侧翼序列，并转换为5×100的二元矩阵作为PlantC2U的输入。该卷积神经网络（CNN）架构包含2个一维卷积层（Conv1D）、1个最大池化层与1个展平层，采用非线性softmax激活函数计算C-to-U RNA编辑的预测概率
图2. 正负样本不同组合下的模型性能评估。(A) 分别以RNS1和RNS2作为阴性样本时，基于马修斯相关系数（Matthews Correlation Coefficient, MCC）的模型性能表现。其中RNS1代表来自原始阴性样本的随机阴性序列，RNS2代表来自不同序列聚类集的随机阴性序列，Ratio代表正负样本比例。(B) 受试者工作特征（Receiver Operating Characteristic, ROC）曲线显示，卷积神经网络（CNN）的性能优于随机森林（RF）、支持向量机（SVM）模型及PREPACT工具。(C) 基于精确召回曲线下面积（Area Under the Precision-Recall Curve, AUC-PRC）的模型性能评估
图3. 红树植物<i>K. obovata</i>质体基因组图谱，包含83个基因、8个核糖体RNA（rRNA）、37个转运RNA（tRNA）及35%的GC含量。IRA与IRB代表反向重复区域，LSC为大单拷贝区域，SSC为小单拷贝区域
图4. <i>K. obovata</i>质体RNA编辑位点的鉴定结果。(A) 不同<i>K. obovata</i>组织中的整体C-to-U RNA编辑水平，数值为三次生物学重复的均值±标准差。通过单因素方差分析（one-way ANOVA）检验RNA编辑水平的显著差异，不同小写字母代表组间差异显著（P < 0.05）。(B) 从<i>K. obovata</i>叶片转录组中鉴定的单核苷酸变异（SNV），柱状图展示每个<i>K. obovata</i>转录组中各SNV类型（如C>T、CT）的占比。(C) 耐寒（CT）与非耐寒（NCT）转录组的整体C-to-U RNA编辑水平箱线图。(D) 耐寒与非耐寒转录组中4个基因（<i>psbA</i>、<i>psbM</i>、<i>rps12</i>与<i>ndhA</i>）的表达水平箱线图。*代表0.01 < P ≤ 0.05，**代表0.001 < P ≤ 0.01。(E) 使用STRING在线工具对4个基因进行的功能富集分析结果。(F) 耐寒与非耐寒转录组中<i>ndhA</i>基因C371与C991位点的C-to-U RNA编辑水平
图5. 质体RNA编辑位点的鉴定与注释工作流程。(A) 包含原始测序数据质量控制、清理后读段比对、RNA编辑位点调用及C-to-U RNA编辑位点注释的分析流程。(B) 从24种红树植物不同组织中鉴定的C-to-U RNA编辑位点数量
图6. 植物质体C-to-U RNA编辑位点在线预测工具PlantC2U的界面

搜集汇总

背景与挑战

背景概述

该数据集专注于植物叶绿体C-to-U RNA编辑的预测，通过卷积神经网络（CNN）深度学习模型分析跨物种序列景观，提供了详细的编辑位点信息和模型性能评估。数据集包括多个表格和图像，涵盖红树林植物K. obovata的RNA编辑位点识别、功能注释以及环境响应分析，并附有在线预测工具PlantC2U的界面，支持RNA-seq数据验证和跨物种应用研究。