1. Introduction

Currently, analytical tools fall into two categories: web-based platforms and locally executed command-line scripts. The former is often user-friendly but typically limited in functionality, requiring users to visit multiple platforms to fully process and analyze omics data. The latter rely on the operation of scripts/codes, which can enhance flexibility but demand computational expertise due to complexities in installation and operation. Therefore, integrating the strengths of both approaches through "one-stop" online platforms offers a solution to advance genomics research. This need is especially acute in plastome (plastid genomic) research, where few integrative analytical platforms are currently available.

Due to its relatively small genome size, conservative gene content, and moderate evolutionary rate, the plastome is an indispensable molecular data source that has been widely applied in phylogenomics, evolutionary biology, comparative genomics, population genetics, phylogeography, and genetic engineering (Wicke et al., 2011; Jansen and Ruhlman, 2012; Straub et al., 2012; Jin and Daniell, 2015; Daniell et al., 2016; Li et al., 2019; Stull et al., 2021). Many specialized tools exist for plastome analyses. Assembly tools include SPAdes (Bankevich et al., 2012), IOGA (Bakker et al., 2016), ORG.Asm (Coissac et al., 2016), NOVOplasty (Dierckxsens et al., 2017), Organelle_PBA (Soorni et al., 2017), chloroExtractor (Ankenbrand et al., 2018), GetOrganelle (Jin et al., 2020), ptGAUL (Zhou et al., 2023), Oatk (Zhou et al., 2024), and TIPPo (Xian et al., 2025). Annotation tools include DOGMA (Wyman et al., 2004), CpGAVAS (Liu et al., 2012), Plann (Huang and Cronk, 2015), Verdant (McKain et al., 2017), GeSeq (Tillich et al., 2017), AGORA (Jung et al., 2018), CPGAVAS2 (Shi et al., 2019), and PGA (Qu et al., 2019). Visualization tools include OGDRAW (Greiner et al., 2019), CGView (Stothard et al., 2019), and CPGView (Liu et al., 2023). Phylogenetic tools include HomBlocks (Bi et al., 2018), ORPA (Bi et al., 2024), and OGU (Wu et al., 2024). Comparative genomic tools include VISTA (Frazer et al., 2004), progressiveMAUVE (Darling et al., 2010), IRscope (Amiryousefi et al., 2018), IRplus (Díez Menéndez et al., 2023), and CPJSdraw (Li et al., 2023). Despite these tools, there are still various technical issues regarding plastome sequence processing and utilization, including standardization of genome structure, evaluation of annotation quality, accurate extraction of target sequences, streamlined pre- and/or post-alignment sequence processing, automatic screening of molecular markers, and gene-level visual comparison of structural variations and annotation completeness. These issues are elaborated below.

First, accurate identification of the quadripartite structure of plastomes remains a challenge. The plastomes of most photosynthetic seed plants are highly conserved, with a quadripartite structure comprising large (LSC) and small (SSC) single-copy regions and two identical inverted repeat (IR) regions (Wicke et al., 2011; Jansen and Ruhlman, 2012). In some cases, two IR copies are not completely identical, due to the imperfections (mismatches) or heterogeneities introduced into genome sequences (Turudić et al., 2022; Díez Menéndez et al., 2023), suggesting there is still room for IR identification by current methods (Turudić et al., 2022). The standardization of quadripartite structure relies on precise IR boundary identification (Turudić et al., 2022). Current main plastome annotation tools, such as the web-based GeSeq (Tillich et al., 2017) and command-line PGA (Qu et al., 2019), incorporate the identification of IR copies, but lack quadripartite structure adjustment. Specialized tools like IRscope (Amiryousefi et al., 2018) and CPJSdraw (Li et al., 2023) focus solely on identifying IR boundary, with CPJSdraw resolving accurate LSC-IRa junctions; however, neither method produces standardized plastome sequences. Tools like NOVOWrap (Wu et al., 2021) and Plastaumatic (Chen et al., 2022) claim to have the functions of adjusting non-standardized form to standardized form. However, the "validate.py" script in NOVOWrap depends on reference sequences for adjustments, risking incorrect SSC orientation when using distantly related or unsuitable references. The "standardize_cpDNA.sh" bash script in Plastaumatic cannot be run without a Linux operating system.

Second, accurate annotation of whole plastomes remains a challenging task, despite the availability of multiple plastome annotation tools, including web tools such as GeSeq (Tillich et al., 2017), CPGAVAS2 (Shi et al., 2019), and Chloe (https://chloe.plastid.org/annotate.html), along the command-line tool PGA (Qu et al., 2019). Annotation principles of existing plastome annotation tools were reviewed and compared in Qu et al. (2023). Critical challenges include annotating pseudogenes, redundant genes, and the trans-splicing rps12 gene. Pseudogenes are non-functional gene fragments derived from functional genes (Goodhead and Darby, 2015), arising from mutations like premature stop codons or frameshifts (Xiao et al., 2016). Their presence complicates genome annotation (Qu et al., 2023), yet accurate identification of these evolutionary remnants is critical for investigating pseudogenization, novel gene origin, and structural rearrangements. However, no tool reliably annotates pseudogenes or evolutionary relics. Trans-splicing, unlike cis-splicing, joins exons derived from either distinct pre-mRNAs or duplicated regions within the same gene. The rps12 gene—typically containing three exons (or occasionally two) distributed across two distinct pre-mRNAs—is prone to misannotation or incomplete annotation, as its fragmented exons make manual exon link particularly challenging (Tillich et al., 2017; Shi et al., 2019; Qu et al., 2023). Currently, only GeSeq and CPGAVAS2 have the ability to annotate rps12, but their accuracy and reliability require comprehensive validation. Although CPGview enables visual verification of rps12 annotation accuracy, a comprehensive assessment of the hit rate and link accuracy of rps12 in angiosperms and broader eukaryotes is lacking.

Third, tools for quantitatively evaluating plastome annotation quality remain underdeveloped. There are tools like BUSCO (Waterhouse et al., 2018) and CEGMA (Parra et al., 2007) that can assess nuclear genome assembly/annotation completeness, equivalent solutions for plastomes are absent. Our previous study proposed three methods for quantitatively evaluating plastome annotations, i.e., gene number comparison, gene length difference comparison, and gene sequence similarity comparison (Qu et al., 2023). However, no dedicated tools exist to implement these assessments. Additionally, gene-level visualization tools for direct comparison of annotation quality of plastomes are still lacking.

Fourth, accurate extraction of coding and noncoding sequences from plastomes is a complicated task. The genic composition of plastomes includes diverse elements: protein-coding (PCG) genes, RNA (tRNA and rRNA) genes, intron-containing vs. intron-lacking genes, cis-splicing vs. trans-splicing genes, non-overlapped/non-nested genes and overlapped/nested genes, single-copy vs. multi-copy genes, and genes spanning start-end position or the start-end position within the noncoding (intergenic/intronic) regions (Raubeson and Jansen, 2005; Qu et al., 2023). Despite the complexity, no dedicated tools offer customizable extraction modes to address these diverse demands.

Fifth, tools for pre- and post-alignment processing remain not integrated, and it is difficult to search customed tools to meet diverse or specialized needs. Although most land plants have conserved plastome structures, many lineages possess highly rearranged plastomes with disordered gene order (Knox, 2014; Mower and Vickrey, 2018; Qu et al., 2022), complicating the use of intergenic regions (Shaw et al., 2007). For plant taxonomists lacking programming expertise, tasks such as assessing codon status in PCGs, identifying missing taxa in sequence datasets, extracting codon sites or sub-alignments from alignments, generating alignments for PartitionFinder, and automatically screening markers based on sequence variation prove challenging. Therefore, the development of streamlined methods for pre- and/or post-alignment sequence processing are urgently needed.

Sixth, tools for micro-synteny comparison of plastomes are lacking. Given the limited number of genes in plastomes (Raubeson and Jansen, 2005), gene-level micro-synteny comparison is realistic and effective. This approach can reveal structural variations such as inversion, translocation, and duplication (Palmer, 1991; Mower and Vickrey, 2018), as well as annotation completeness issues such as gene loss/gain (Qu et al., 2023). Visualizing pairwise gene arrangements could further evaluate annotation quality at the gene level. However, specific tool remains underdeveloped for performing plastome-specific micro-synteny comparisons.

To address these critical challenges in plastome analysis, we developed PlastidHub, an online platform integrating an interconnected task-based user interface for end-to-end processing of plastome sequences. All tools in PlastidHub possess batch processing capability, critical for handling the large-scale datasets routinely analyzed by botanists and evolutionary biologists.

2. Materials and methods 2.1. Implementation

PlastidHub is a web application developed with Java, TypeScript, and Perl, designed to cater to users who favor browser-driven workflows or those lack familiarity with UNIX command-line tools. The front-end web interface was developed using vue3 (https://cn.vuejs.org/) and Element-Plus (https://element-plus.org/). PlastidHub is compatible with recently released modern browsers, including Google Chrome, Microsoft Edge, Mozilla Firefox, Opera, Apple Safari, 360 Speed Browser X, 360 Secure Browser, QQ Browser, and Huawei Browser. The back-end web framework uses SpringBoot v.3.3.2 (https://docs.spring.io/spring-boot/index.html) with RabbitMQ (https://www.rabbitmq.com/) as message queue and Redis (https://redis.io/) storing result status, which can manage data and parameters submitted by users, and then relay the analysis results (generated files) to the front-end interface for users to manage (batch selection, download, and deletion). The Perl-based (v.5.26.1) back-end architecture of PlastidHub supports execution across Windows, Linux, and Mac via command-line interfaces.

2.2. Analysis workflow

PlastidHub is an integrated plastome analysis platform including 10 tool kits: annotation, assessment, submission, visualization, extraction, pre-alignment, post-alignment, phylogeny, barcoding, genomics, each designed for specific tasks. Users can dynamically chain these toolkits to build customized workflows. For examples, "annotation → assessment"; "annotation → assessment → submission"; "annotation → assessment → visualization"; "annotation → assessment → extraction → pre-alignment → (post)-alignment → phylogeny"; "annotation → assessment → extraction → pre-alignment → (post)-alignment → barcoding"; and "annotation → assessment → plastomics". Workflows are fully flexible, users can start or terminate analyses at any stage.

2.3. Supported file formats

PlastidHub supports GenBank (.gb/.gbf/.gbk), FASTA (.fasta/.fa/.fas/.fsa), and TEXT (.txt) input files, as specified in the File Upload Panel of each tool. The format of output files is user-selectable and include GenBank, FASTA, BED (.bed), TABLE (.tbl), TAB-delimited (.txt), SVG (.svg), PNG (.png), Newick TREE (.tre), and CONFIG (.conf).

2.4. File number and size limitation

The file number and size allowed by PlastidHub are displayed in the static pop-up prompt box of each tool.

2.5. Special characters in filenames

Some functionalities in PlastidHub rely on third-party tools, so the use of non-English characters and certain special characters (~! @#$%^ & *()+ = []}{; : '" < >, /?|) is not recommended. In particular, we recommend to use underscores (_) or dots (.) instead of spaces () or hyphens (−).

2.6. Help documents and notes

To facilitate use of this new platform, PlastidHub provides detailed user guides, test example files, and warning prompts when entering unreasonable parameter values. Help documents can be easily acquired from the document pages corresponding with each tool. Structured help documents can also be accessed in the supplemental information. Notes and warning prompts for each tool are shown in the tool interface. In general, use of unreasonable parameter values will prompt an error warning encouraging use of values within a specified range. Warnings will also appear in cases where files are incorrectly uploaded (e.g., due to wrong file format) or due to restrictions in file number or size. These resources streamline platform adaptation, enabling researchers to focus on biological insights.

2.7. Test data and tool evaluation and comparisons

To evaluate the performance of newly developed tools in PlastidHub, 15, 206 eukaryotic plastome sequences were retrieved from the NCBI RefSeq database (https://ftp.ncbi.nlm.nih.gov/refseq/release/plastid/, accessed 2024.5.16). First, we count the proportion of non-standardized quadripartite structure across all 15, 206 eukaryotic plastome sequences using the Quadripartition tool. Second, we evaluated the hit rate and link accuracy of rps12 with the plastome annotation tool PGA v.2.0 across 15, 206 eukaryotic and 13, 731 angiosperm plastome sequences. To evaluate plastome annotation consistency of PGA v.2.0, the number of annotated genes in GenBank and PGA annotations were compared for 13, 731 angiosperm plastid sequences. Third, we applied the tools of "Assess Gene Number" and "Assess Gene Length" to evaluate the annotation quality of the target species Rosa roxburghii using Amborella trichopoda as reference. Fourth, to evaluate the accuracy of the plastome Extraction tool, the number of extracted genes and PGA-annotated genes in the 13, 731 angiosperm plastid sequences was compared. Fifth, to evaluate the effectiveness of the Barcoding tool in automatically screening molecular markers, plastomes from Amaranthaceae species including Amaranthus caudatus, A. tricolor, Atriplex centralasiatica, Chenopodium album, C. quinoa, Salicornia bigelovii, S. europaea, Spinacia oleracea, Suaeda glauca, and S. salsa were used. The choice of taxonomic level (e.g., order, family, genus, species, or mixed) does not affect barcoding results, as the final barcoding results are sorted by variable sites and parsimony informative sites—metrics that remain phylogenetically meaningful across all taxonomic scales. Thus, selection criteria remain consistent, regardless of whether the taxa are closely related or broadly sampled. Sixth, to evaluate the functionality (showing micro-synteny structural variations and evaluating annotation completeness) of the Gene Homology tool, we conducted plastome comparisons of Amborella trichopoda against other angiosperms (Rosa roxburghii and Salicornia bigelovii) and a gymnosperm species (Calocedrus decurrens), as well as a comparison of the gymnosperm species C. decurrens and C. rupestris.

We conducted a series of comparisons with existing tools. For "Main Function 1", we compared the consistency of quadripartite standardization between the Quadripartition tool in PlastidHub and the "validate.py" tool in NOVOWrap using 53 angiosperm plastomes from 48 orders. For "Main Function 2", we have compared PGA v.1.0 with GeSeq using practical examples in our previous published paper (Qu et al., 2019). For further comparison of the newly developed function in PGA v.2.0, we use above mentioned 53 angiosperm plastomes to compare the hit rate and link accuracy of rps12 for PGA v.2.0 with GeSeq and CPGAVAS2. For "Main Function 4", to demonstrate the accuracy and flexibility of the Extraction tool in PlastidHub, we use above mentioned 53 angiosperm plastomes to compare the accuracy of sequence extraction between PlastidHub and PhyloSuite. For "Main Function 3/5/6", they are completely novel tools, without other existing tools to compare. Therefore, we use true data examples to demonstrate how these three tools combine simplicity and ease of use with accuracy and practicality in evaluating annotation quality, assisting primer design, and displaying micro-synteny structural variations.

For some PlastidHub tools that are similar to other existing popular tools, we have introduced them in the paragraph titled "Similar functions with other tools" within the Results and Discussion section, allowing users to choose their own preferences. For completing the streamlining analysis of plastid phylogenomics and comparative genomics, we have developed some personalized new functions that are not found in other published papers and introduced them in the paragraph titled "Streamlining analysis with personalized new functions in three toolkits: Pre-Alignment, (Post)-Alignment, and Phylogeny".

2.8. Performance test

PlastidHub's scalability was benchmarked under simulated workloads, evaluating platform behavior across three escalating scenarios: light (20 users), moderate (60 users), and heavy (100 users) concurrent loads. In each scenario, virtual users randomly executed diverse tasks (e.g., annotation, extraction), submitting requests within a 10-s window to mimic real-world usage patterns.

3. Results and discussion 3.1. Interrelated user interface design of PlastidHub

To create an easy-to-use web-based tool kit for researchers engaged in phylogenomic and comparative genomic analyses using plastome data, we generated task-based interfaces where the users can easily choose a tool to accomplish their analysis goals (Fig. 1). PlastidHub contains three interfaces, i.e., "P-Utils" (tool interface), "Document" (help interface), and "My Files" (download interface).

For the "P-Utils" tool interface, the left panel provides links to task-running interfaces of 25 tools across 10 tool kits. The right panel displays a comprehensive list of all 25 tools in PlastidHub, enabling users to instantly navigate to any specific tool. Each tool has a brief introduction to help users to make a right choice. Within each tool's interface, there are interactive options for uploading input files (required), selecting parameter values (optional), submitting (required) or resetting (optional) the task, and linking to the help document for reference. To avoid submission errors, PlastidHub enforces file quantity limits, specified input formats (with permitted suffixes), and restricts parameters to predefined ranges. If the input files are omitted, the submit button remains disable, triggering a brief warning prompt (e.g., Please upload reference/target file first) for 3 s.

For the "Document" help interface, the left panel shows the links to toolkit-specific help documents for 10 tool kits. The right panel lists "Get Started", "User Guide", "Browser Compatibility", "Tech Stack", and "Content". For the "Content" section, a hyperlink-based help document for each of 25 tools is listed. Within the help document of each tool, paragraph navigation links such as "General Introduction", "Features", "Command-line Tool", and "Example" are set up to help users to quickly locate the target paragraph.

An important customization feature of PlastidHub is the "My Files" download interface, allowing users to check the running status of the submitted tasks and download or delete the completed tasks at any time. Intermediate results are also provided to assist users to conduct custom post-processing. The validity period of the result files stored in the server database is one week. If the download button is clicked, a prompt such as "The result of output files will be stored for a week since task finished" will be displayed. Similarly, if the delete button is clicked, a warning prompt such as "Confirm that you are deleting the result(s)" will be displayed. Each tool kit has a corresponding download interface, and the tool interface is located below the corresponding download interface, making it convenient for users to access directly. It indicates that the tool interface can be freely switched not only with the help interface, but also with the download interface. This design enhances the flexibility of running tasks in PlastidHub.

The middle panel of the home page contains user access and usage statistics, including "Task Handled", "Page Views", and "Unique Visitors", providing real-time feedback on user visits and completed tasks. The bottom panel of the home page contains a "Links" section, which includes "GitHub" and "Change Log", and a "Community" section, which includes "Feedback", "Get Involved", and "Contact", as well as a "Policy" section, which includes "Terms of Use" and "Privacy Policy".

3.2. Functionalities of 25 tools within 10 designed tool kits

Full methodological details, use cases, and workflows for each tool are comprehensively documented in the supplementary file.

3.3. Main function 1: High proportion of non-standardized quadripartite structure of plastomes in NCBI RefSeq database identified by the Quadripartition tool

The quadripartition tool in PlastidHub (Fig. 2A) is newly developed based on an optimized IR-copy-identification method from PGA v.1.0 (Qu et al., 2019). Flexibility is our main consideration for adjusting quadripartition structure. Therefore, we allow the IRb and IRa to be set as completely (100%) or almost (99%) identical, the minimum allowed IR length to be changed manually (e.g., ≥ 1000 bp or ≥ 100 bp), the orientation of the whole plastome sequences or the SSC sequences to be reverse complemented or not, and the standardized plastome sequences as well as coordinates and sequences of LSC, IRb, SSC, and IRa to be outputted. This adaptability enables precise structural standardization, critical for resolving inconsistencies in RefSeq plastome annotations.

The consistency of quadripartite standardization was evaluated by comparing the Quadripartition tool in PlastidHub and the validate.py tool in NOVOWrap (Fig. 2B and Table S1). For 53 angiosperm plastomes, a total of 60 unique IRs were identified across four runs (two PlastidHub runs and two NOVOWrap runs), indicating inconsistent identification of IRs within and between tools. Specifically, there is 1 unique IR in each of PlastidHub run1/2, but 3 missing IRs and 3 unique IRs were identified in NOVOWrap run1/2, respectively. For PlastidHub run2, a total of 10 IRs with warnings "Nucleotide and/or length of two IR copies may not be exactly the same!" indicated different IRs that need to be checked (Table S1). Among these 10 different IRs, 4 are corrected to identical IRs by NOVOWrap run2, but 6 are failed to be corrected by NOVOWrap run2, indicating the inconsistency of NOVOWrap in IR identification. For 1 identical IR in PlastidHub run1/2 (Pandanales_Pandanaceae_Pandanus_tectorius_NC_042747), NOVOWrap run1/2 identified it as a different IR. For the IR in a long plastid sequence (Vitales_Vitaceae_Vitis_romanetii_NC_056348), PlastidHub run1 identified a long and different IR containing four rRNA copies in two IR copies, while PlastidHub run2 identified a canonical and different IR containing two rRNA copies in two IR copies. NOVOWrap run1 did not identify IR in this long plastid sequence, but NOVOWrap run2 identified a long and identical IR. These results demonstrate the Quadripartition tool in PlastidHub tends to consistently tolerate sequence differences between IRb and IRa, while the validate.py tool in NOVOWrap tends to become inconsistency in IR identification but its tolerance threshold is unknown.

The proportion of non-standardized quadripartite structures were counted in 15, 206 eukaryotic plastomes (Fig. 2C). If we set IR ≥ 1000 bp, except 9 abnormal plastomes with run fault and null result, there are 933 plastomes with IR < 1000 bp or no IR and 14, 264 plastomes (93.9% hit rate) with IR ≥ 1000 bp. For the 14, 264 plastomes, there are 10, 210 plastomes (71.6% accuracy rate) with correct quadripartite structure and 4054 plastomes with incorrect quadripartite structures. If we set IR ≥ 100 bp and IR < 1000 bp, except 46 abnormal plastomes with null result, there are 582 plastomes with IR < 100 bp or no IR and 305 plastomes (34.4% hit rate) with IR ≥ 100 bp and IR < 1000 bp. For the 305 plastomes, there are 8 plastomes (2.6% accuracy rate) with the correct quadripartite structure and 297 plastomes with the incorrect quadripartite structure. In total, if we set IR ≥ 100 bp, except 54 abnormal plastomes with run fault and null result, there are 582 plastomes with IR < 100 bp or no IR and 14, 569 plastomes (96.2% hit rate) with IR ≥ 100 bp. For the 14, 569 plastomes, there are 10, 218 plastomes (70.1% accuracy rate) with the correct quadripartite structure and 4351 plastomes with the incorrect quadripartite structure. These results confirm a high prevalence of structural non-standardizations in NCBI RefSeq, and the incorrect start-end position of circular plastomes is primarily caused by the offset of LSC-IRa junction (Turudić et al., 2022).

Quadripartite standardization hinges on the identification of inverted repeats (IR). From the perspective of bioinformatics, identifying IRs can be formulated as identifying the two longest reverse-complements in circular plastomes. While tools for repeat detection abound, Turudić et al. (2022) demonstrated that IR identification accuracy varied across six annotation tools due to pervasive imperfections in plastome assemblies (e.g., sequencing errors, assembly artifacts, or genomic heterogeneity). These imperfections necessitate tolerance for IR mismatches, as identical IRs may reflect artificial homogeneity, while divergent IRs may represent biological reality. Therefore, the newly developed Quadripartition tool in PlastidHub mainly considered consistency in IR identification. When discrepancies arise from nucleotide mismatches or length variations between IRb and IRa, customizable warnings (e.g., "Nucleotide and length of IRb and IRa may not be exactly the same!", "Nucleotide of IRb and IRa may not be exactly the same!", or "Length of IRb and IRa may not be exactly the same!") are logged in warning.txt for user review. This transparency empowers researchers to manually validate IR boundaries and mitigate annotation errors stemming from assembly flaws.

3.4. Main function 2: Flexibility and consistency of the Annotation tool PGA v.2.0

We updated PGA v.1.0 to PGA v.2.0 by providing flexible options to annotate evolutionary remnants, namely pseudogenes and/or redundant genes (Fig. 3A). In some cases, it is difficult to distinguish pseudogenes from redundant genes, especially in highly divergent plastomes (Qu et al., 2023). PGA v.2.0 excels at detecting pseudogene remnants, which are critical markers for tracing the evolutionary history of these fast-evolving genes, especially highly degraded plastomes in heterotrophic plants (Goodhead and Darby, 2015; Hsu et al., 2016; Xiao et al., 2016). For regular phylogenetic analysis, we suggest to exclude annotations of redundant genes/pseudogenes to minimize alignment artifacts that compromise phylogenetic accuracy. For comparative genomics and RNA-editing studies, redundant genes/pseudogenes provide a chance to evaluate functional gene decay under relaxed selection.

PGA v.2.0 provided a flexible parameter to link or unlink rps12, addressing diverse annotation needs. The rps12 gene is a trans-splicing gene, containing three exons across two pre-mRNAs, including exon 1 in LSC, and exon 2 and exon 3 in IR (Fig. 3B). Accurate rps12 gene annotation remains challenging due to fragmented exons (Tillich et al., 2017; Shi et al., 2019; Qu et al., 2023). We compared PGA v.2.0 with GeSeq and CPGAVAS2 for rps12 annotation accuracy using the aforementioned 53 angiosperm species (Fig. 3C and Table S2). Using plastome of Amborella trichopoda as reference, PGA v.2.0 and GeSeq successfully linked rps12 in 51 of 53 species (Fig. 3C and Table S2). In contract, CPGAVAS2 achieved 28 of 53 and 5 of 53 successes in rps12 annotation using Arabidopsis thaliana and 2544 plastomes as reference, respectively. When no external reference sequence being specified, GeSeq successfully linked rps12 in 22 of 53 species. Specifically, rps12 in the long plastid sequence (Vitales_Vitaceae_Vitis_romanetii_NC_056348) was not successfully linked by PGA v.2.0, GeSeq, and CPGAVAS2 (Table S2). The rps12 with a 1-nucleotide intron (Zygophyllales_Zygophyllaceae_Zygophyllum_xanthoxylon_NC_052769) was successfully linked by PGA v.2.0 and CPGAVAS2, highlighting enhanced sensitivity to microstructural variations. CPGAVAS2 primarily failed due to missing exon3 annotations, and GeSeq (no external reference) simultaneously exhibited two annotations with incorrect boundaries for blatX annotations and correct boundaries for Chloe annotations (Table S2). Therefore, we recommend PGA v.2.0 and GeSeq for reliable rps12 annotation, prioritizing reference-guided workflows to minimize errors.

To evaluate rps12 annotation accuracy by at large scale, 15, 206 eukaryotic plastid plastomes from NCBI RefSeq were re-annotated using PGA v.2.0 with default parameter values (Fig. 3D). Using Amborella trichopoda as reference, 15, 200 plastomes were successfully annotated, while 6 from Chlorophyta failed. For 15, 200 eukaryotic plastomes, 14, 191 (93.4% hit rate) contained rps12 annotations with 93–435 bp length variation, while 1009 lacked rps12. For 14, 191 eukaryotic plastomes, 13, 494 (95.1% link accuracy) had full-length rps12 (339–435 bp), while 697 were only annotated with one exon (93–270 bp). For 13, 731 angiosperm plastomes, 13, 119 (95.5% hit rate) contained rps12 (93–435 bp), while 612 lacked rps12. For 13, 119 angiosperm plastomes, 12, 845 (97.9% link accuracy) had full-length rps12 (339–435 bp), while 274 sequences were only annotated with one exon (93–270 bp). These results confirm the robustness of PGA v.2.0, achieving high rps12 detection and link accuracy even when annotating divergent eukaryotic plastomes with angiosperm-specific references.

To evaluate annotation flexibility and consistency, the number of annotated genes in GenBank and PGA v.2.0 were compared using the 13, 731 angiosperm plastomes from NCBI RefSeq (Fig. 3E). GenBank annotations (original submissions) and PGA v.2.0 annotations (re-annotated with Amborella trichopoda as reference) were compared. The order of accession numbers for the angiosperm plastomes from PGA and GenBank was the same between ① and ② in Fig. 3E, as well as between ③ and ④ in Fig. 3E, ensuring direct comparability of annotation quality. Only 224 plastomes had gene numbers fewer than 100 in PGA annotations (Fig. 3E①), while 759 plastomes in GenBank annotations (Fig. 3E③). By comparing ② with ① in Fig. 3E, we found that the plastomes with gene numbers under 100 were scattered across taxa in GenBank annotations, reflecting inconsistent annotation quality. In contrast, by comparing ④ with ③ in Fig. 3E, we found that the plastid sequences with gene numbers under 100 in PGA annotations clustered in a fixed range, indicating systematic annotation rigor. Importantly, all PGA low-gene plastomes were subsets of GenBank low-gene plastomes, suggesting shared sequence features (e.g., fragmentation, pseudogenization) rather than annotation errors. This contrasts with GenBank's inconsistent annotation quality, likely reflecting heterogeneous standards across decades of plastome sequencing, assembly, and annotation practices. Despite RefSeq's curation, data quality variation persists due to evolving methodologies. PGA v.2.0's standardized workflow mitigates this, delivering highly consistent annotations for comparative studies.

3.5. Main function 3: Evaluating plastome annotation quality with the Assessment tool

We proposed two quantitative methods to evaluate the annotation quality of plastomes: gene number completeness and gene length accuracy (Qu et al., 2023). Using plastome of Amborella trichopoda as a reference, we evaluated the annotation quality of Rosa roxburghii (Fig. 4). For gene number assessment, the annotation completeness of genes can be judged by number of hitting (nHG), missing (nMG), and redundant (nRG) genes; percentage of hitting (pHG), missing (pMG), and redundant (pRG) genes; number of hitting (nHP, nHT, nHR), missing (nMP, nMT, nMR), and redundant (nRP, nRT, nRR) PCGs, tRNAs, and rRNAs; and names of hitting, missing, and redundant PCGs, tRNAs, and rRNAs (Fig. 4A). Since plastomes do not contain many genes, it would be reasonable for users to manually check annotations (hitting, missing, and redundant) of each gene to determine biological gene loss from annotation errors or pseudogene misclassification. For gene length assessment, length differences between target and reference genes are ranked by magnitude (Fig. 4B), and those with the largest differences (or those larger than a specified threshold) can be checked. Gene lengths of tRNAs and rRNAs are more conserved than those of PCGs across plant species, with rare exceptions, so the length differences of PCGs, and tRNAs and rRNAs are separately compared. These tools empower users to refine gene annotations by identifying missing/poorly bounded genes and redundant false positives, ensuring high-quality plastome data.

3.6. Main function 4: Accuracy of the plastome Extraction tool

Extracting coding and noncoding sequences from plastomes is an important step for downstream analyses of selection tests or phylogenetic reconstructions. Given the circular structure and diverse gene types of plastomes, accurate extraction of all target sequences is not trivial. Our newly developed extraction tool in PlastidHub accommodates diverse extraction patterns (Fig. 5), including canonical genes with or without introns, and specific nested genes (Fig. 5A), trans-splicing genes (Fig. 5B), and genes avoiding (Fig. 5C) or spanning (Fig. 5D) the start-end position of circular plastome. In addition to PCGs with introns and sequences among PCGs, genes with introns and intergenic sequences are extracted. Furthermore, coding sequences with linked and unlinked CDSs and tRNAs are considered, as well as noncoding sequences. Customizable parameters enable tailored extraction, ensuring high accuracy for complex plastome structures.

To compare the Extraction tool in PlastidHub and the "Extract GenBank file" (EGBF) tool in PhyloSuite, we used aforementioned 53 angiosperm plastomes to evaluate the number and accuracy of different types of extracted sequences (gene, CDS, tRNA, rRNA, and noncoding regions) (Fig. 5E and Table S3). For gene, CDS, tRNA, and rRNA, PlastidHub and PhyloSuite extracted equal total number of target sequences (Fig. 5E and Table S3). Among the 53 species, the trnH-GUG gene in 8 species spans the start-end position, PhyloSuite extracted two partial sequences, while PlastidHub retrieved complete trnH-GUG. PhyloSuite failed to link exons of unlinked rps12 exons, whereas PlastidHub output complete rps12 regardless of link status. These results confirm PlastidHub's superior handling of circular plastome complexities, ensuring accurate sequence retrieval for downstream analyses.

It is worth mentioning that the number of noncoding sequences extracted by PhyloSuite (N_avg. = 109.0) is much smaller than that of PlastidHub (N_avg. = 132.2). Notably, PlastidHub correctly extracted all noncoding sequences, while PhyloSuite generated an average of 4.1 erroneous noncoding sequences per species across the 53 analyzed plastomes (Fig. 5E and Table S3). This discrepancy arises because PhyloSuite fails to extract intron sequences unless they are explicitly labeled with the "intron" feature, resulting in the omission of 20 introns from 18 intron-containing genes. Additionally, PhyloSuite cannot handle nested genes well, such as matK-trnK-UUU, further contributing to incomplete noncoding sequence extraction. Similarly, Geneious, a widely used commercial platform, lacks functionality for extracting intergenic regions or noncoding sequences. These limitations highlight that PhyloSuite or Geneious are not optimized for comprehensive noncoding sequence extraction from plastomes, underscoring the need for tools like PlastidHub, which is specifically designed for this purpose.

GenBank file permits features (e.g., gene, CDS, tRNA, rRNA, etc.) to be listed in any order, as long as the provided coordinates are accurate. PhyloSuite requires features to follow physical gene order for correct intergenic region extraction. Disordered features in GenBank files lead to pervasive extraction errors in PhyloSuite. PlastidHub resolves this limitation, accurately processing both ordered and disordered GenBank features to ensure precise sequence extraction.

To validate the scalability of the Extraction tool in PlastidHub, the gene numbers from PGA v.2.0 annotations were analyzed across 13, 731 angiosperm plastomes (Fig. 5F). Re-annotation with Amborella trichopoda as reference, gene numbers concentrated at ~110, aligning with typical gene numbers in angiosperm plastomes (Wicke et al., 2011; Jansen and Ruhlman, 2012). The hit rate of gene extraction from PGA annotation was used to evaluate the accuracy of the extraction tool in PlastidHub (Fig. 5G). A strong positive correlation (Pearson correlation: R = 1, p < 2.2e-16) was observed between extracted and annotated gene counts, with ≤ 1-gene discrepancies in most cases (Fig. 5G). These results affirm the extraction tool in PlastidHub as a state-of-the-art solution for retrieving canonical and special coding and noncoding sequences from plastomes, adaptable to diverse research needs.

3.7. Main function 5: Automatic screening of molecular markers using the Barcoding tool

DNA barcodes is pivotal for species identification, phylogenetics, population genetics, phylogeography, conservation genetics, and biodiversity studies, but automated plastid marker screening remains challenging.

We developed a novel barcoding tool to address this issue by enabling batch calculation and extraction of variable, invariable, and gap sites from sequence alignments (Fig. 6). This tool streamlines the screening of molecular markers by integrating variable sites, parsimony informative sites, and total alignment length into a unified workflow, eliminating the need for third-party tools and accelerating downstream applications and analyses. Specifically, both the number and percentage of variable sites/parsimony informative sites within each alignment are calculated. In addition, both variable sites/parsimony informative sites and their positional data of nucleotides within each alignment are extracted. Most importantly, the alignments can be sorted by the percentage of variable or parsimony informative sites to prioritize informative markers (Fig. 6A–H). Notably, variable or parsimony informative sites across PCGs, RNAs, introns, and intergenic regions are separately sorted and categorized, with total alignment length incorporated as a criterion for marker selection (Fig. 6A–H), a critical factor in primer design (Ye et al., 2012). Boxplot analyses revealed that intergenic regions, introns, and PCGs exhibit significantly higher variability in both variable and parsimony informative sites compared to RNAs, tRNAs, and rRNAs (Fig. 6I), reflecting differences in evolutionary rates among genomic regions (Wolfe et al., 1987; Magee et al., 2010).

3.8. Main function 6: Micro-synteny structural comparison using the Gene Homology tool

Whole-plastome alignment typically highlights macro-scale structural variations. mVISTA (Frazer et al., 2004) is one of the most widely used tools for plastome comparison. However, its utility is hampered by the technical complexity of generating input files, which poses a significant barrier for researchers without bioinformatics expertise. There are other tools available for comparative analysis of plastomes, such as progressiveMAUVE (Darling et al., 2010), which are widely applied in studies focusing on plastomes with structural variations. However, tools capable of performing micro-synteny analysis in plastomes remain scarce.

We developed a novel tool "gene homology" to address this gap by generating pairwise circular maps of two plastomes that connect homologous genes and mark non-shared genes (Fig. 7). This tool enables pairwise visual comparison of each plastid gene, ultimately displaying micro-synteny structural variations and evaluating gene-level annotation completeness. It not only identifies structural variations such as inversions, translocations, and duplications, but also quantifies the extent of annotation completeness by accounting for gene losses/gains. Regardless of whether the reference and target plastomes have IR copies, including scenarios with two IRs or a single IR, this tool accommodates all structural possibilities (Fig. 7A). To demonstrate the flexibility of this tool, we compared plastome of Amborella trichopoda with other angiosperms (Rosa roxburghii, Salicornia bigelovii) and a gymnosperm species (Calocedrus decurrens), as well as gymnosperm species pair Calocedrus decurrens vs. Calocedrus rupestris (Fig. 7B). As far as we know, this is the first tool for pairwise visualization of plastomes at the gene level, offering unprecedented resolution for micro-synteny structural variation analysis.

3.9. Similar functions with other tools: Submission and Visualization

As far as we know, there are two tools that can convert GenBank flatfiles to the tbl-formatted files required for NCBI/EMBL/DDBJ database submission: gbf2tbl.pl (NCBI) and GeSeq (Tillich et al., 2017). There are six tools that can perform plastome visualization: OGDRAW v.1.3.1 (Greiner et al., 2019), CGView (Stothard et al., 2019), CPGAVAS2 (Shi et al., 2019), Chloroplot (Zheng et al., 2020), Chloe (https://chloe.plastid.org/annotate.html), and CPGView (Liu et al., 2023). GeSeq is the only tool that can conduct both format conversion and gene visualization of annotated plastomes. PlastidHub complements these tools by providing user-friendly submission workflow and enhanced visualization capabilities. Its submission tool can batchly generate tbl-formatted files with 100% format compliance (validated on 200+ submissions), including mandatory fields like "gene", "CDS", "tRNA", and "rRNA", supporting multi-genome submissions. The submission tool is capable of recognizing rps12 exon-linking algorithm, achieving 100% accuracy in resolving the rps12 split issue (demonstrated across 200+ plastomes), thereby eliminating the need for manual corrections previously required for NCBI submissions. Its visualization tool uniquely shows linked dispersed repeats (forward and inverted), with different repeats including minimum allowed percent identity, minimum allowed length, and BLAST e-value to refine repeat annotations. This flexibility supports tailored analysis of structural variation, setting PlastidHub apart as a multifunctional platform for plastome curation.

3.10. Streamlining analysis with personalized new functions in three toolkits: Pre-alignment, (Post)-Alignment, and Phylogeny

PlastidHub introduces five customizable tools to optimize plastid phylogenomic workflows: "sort flanking genes", "check bad codons", "check missing taxa", "extract sub-alignments", and "RAxML gene tree". The tool "sort flanking genes" was specifically developed for plastomes with significant structural rearrangements, such as inversions and translocations. This is because it is essential to adjust the orientations of intergenic regions (to a uniform direction) before alignment and phylogenetic reconstruction. This tool improves the alignment accuracy for intergenic regions in plastomes, and can also trace the evolutionary origins of gene remnants in intergenic regions, catering to users with specialized analytical requirements. The tool "check bad codons" detects problematic codons in batch-submitted of PCG matrices, allowing users to filter out low-quality sequences to improve alignment accuracy. Since reliable PCG matrices exclude internal stop codons, these codons serve as indicators of potential errors in PCG annotations. The tool "check missing taxa" identifies taxa absent in batch-submitted sequence matrices, generating a matrix that displays species and sequences coverage to help users to assess dataset integrity and detect potential false-positive issues linked to specific species or plastid regions. The tool "extract sub-alignments" accepts a user-defining species list to create sub-alignments from batch sequence submissions, while pruning gap-only sites in the sub-alignments. This tool is important for illuminating how species sampling influences phylogenetic results. The tool "RAxML gene tree" conducts ML phylogenetic inference using RAxML on batch sequence alignments, enabling users to infer gene and species trees for assessing phylogenetic conflicts.

3.11. Response time, throughput and scalability 3.11.1. Response time analysis

Under a load of 20 concurrent users, PlastidHub achieved an average response time of 295 ms (optimal performance) (Fig. 8A). At 60 users, response time rose to 442 ms (moderate increase), and at 100 users, reached 619 ms (sub-second response). All requests under peak load completed in < 1 s, demonstrating robust responsiveness and scalability under growing demand.

3.11.2. Throughput analysis

The throughput of PlastidHub increases linearly with the number of concurrent users, from 3.84 requests/second at 20 users to 11.40 requests/second at 100 users (Fig. 8B). This near-linear scalability indicates a well-designed architecture that efficiently utilizes server resources to handle increasing loads. Notably, PlastidHub maintained a 0% error rate across all loads, with no failed or timed-out requests even at peak load (Fig. 8C), confirming exceptional stability.

3.11.3. Scalability analysis

When response times or processing capacity approach bottlenecks, PlastidHub can quickly scale horizontally by adding more servers. Horizontal expansion ensures sustained performance. This elastic scaling reduces latency and multiplies task-processing capacity on demand.

Finally, the performance test results demonstrate that PlastidHub exhibits exceptional concurrent processing capability and robust scalability. Specifically, the system can: (1) stably handle high loads of up to 100 concurrent users; (2) maintain all request response times within acceptable thresholds; (3) achieve a 0% error rate across varying load conditions; (4) scale throughput linearly with added computational resources. These results confirm PlastidHub's reliability, operational efficiency, and readiness for deployment in production environments with demanding workloads.

4. Conclusions

While tools like GeSeq (Tillich et al., 2017) possess partial functionalities with PlastidHub, none match its breadth of customizable functions. PlastidHub currently offers a set of 25 tools spanning diverse plastome analyses, with 19/25 tools operating independently of third-party tools. The remaining six integrate BLAST + v.2.16.0 (Camacho et al., 2009), Circos v.0.66 (Krzywinski et al., 2009), MAFFT v.7.3 (Katoh and Standley, 2013), and RAxML v.8.2.10 (Stamatakis, 2014). PlastidHub will be continuously updated, adding new tools and incorporating emerging methods to serve the expanding needs of plastome research.

PlastidHub's cloud-based architecture eliminates the need for software installation/updates and high-performance computing investments. User experience drives platform development, exemplified by seamless switching between analysis, help, and download interfaces during workflow execution. User feedback and field advancements will guide daily updates and tool expansions, ensuring that PlastidHub remains at the forefront of plastome informatics.

Acknowledgements

We thank Dan Zou, Yu-Liang Chen, Zhen-Xiang Feng, Yuan-Ting Gu, and Rui-Yu Zhang for testing the PlastidHub. We thank the editors and reviewers for constructive comments and suggestions on improving this study. This study was funded by the Natural Science Foundation of Shandong Province (ZR2020QC022), the Science and Technology Basic Resources Investigation Program of China (No. 2019FY100900), the Major Program for Basic Research Project of Yunnan Province (202401BC070001), Yunnan Revitalization Talent Support Program: Yunling Scholar Project to Tingshuang Yi, and the open research project of "Cross Cooperative Team" of the Germplasm Bank of Wild Species, Kunming Institute of Botany, Chinese Academy of Sciences.

CRediT authorship contribution statement

Na-Na Zhang: Writing – original draft, Visualization, Formal analysis, Data curation. Gregory W. Stull: Writing – review & editing, Supervision, Investigation. Xue-Jie Zhang: Writing – review & editing, Visualization. Shou-Jin Fan: Writing – review & editing, Supervision, Conceptualization. Ting-Shuang Yi: Writing – review & editing, Writing – original draft, Supervision, Funding acquisition, Conceptualization. Xiao-Jian Qu: Writing – review & editing, Writing – original draft, Visualization, Supervision, Funding acquisition, Formal analysis, Conceptualization.

Data accessibility statement

All relevant data and code can be found within the article and on the PlastidHub website (https://www.plastidhub.com/) and GitHub (https://github.com/quxiaojian/PlastidHub).

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.pld.2025.05.005.