{"id":"098e228e-d800-4ad7-85a0-11bad40f3b14","arxiv_id":"2510.10950","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"ChloroScan recovers plastid genome bins from metagenomes using a deep-learning contig filter and marker-gene-guided binning, outperforming MetaBAT2 on simulated marine data.","lead":"ChloroScan is a new automated pipeline that separates algal plastid genomes from mixed seawater DNA and assembles them into draft genomes. It recovers more complete plastid genome bins than the standard MetaBAT2 binner in simulated tests and pulled 16 such genomes from four ocean samples.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Real-data ptMAG quality may be inflated: the same plastid marker database guides binning and defines completeness/purity; split-marker cross-validation is needed.","rationale":"The reader's conditional verdict already flags several issues, including the partly circular marker use, so my concern does not move the verdict. I rate it as the single most load-bearing concern because it targets the numerical 16/70/90 real-data claim directly, whereas the reader's stated weakest assumption (Corgi/marker database coverage of target lineages) is an acknowledged limitation and does not affect the simulated MetaBAT2 comparison. The paper has real independent support: deposited code/data, a controlled simulation with ground-truth AMBER evaluation, and transparent discussion of the dinoflagellate failure. The split-marker test is feasible with the deposited reproducibility repository and would settle whether the real-data quality estimates are trustworthy.","tokens_in":16372,"tokens_out":6592,"duration_ms":62720,"concrete_test":"Randomly split the custom plastid marker database into two disjoint halves, A and B. Rerun ChloroScan on the four Tara samples with binny guided by A only and compute completeness/purity from B only, then repeat with roles reversed. If the 16 bins meeting 70/90 under the same-database QC largely disappear or shift under cross-validated marker sets, the real-data quality is at least partly a self-consistency artifact; if all 16 remain above thresholds, the concern is resolved. A supporting check on the deposited CAMISIM simulations: compare each bin's reported marker-based completeness with AMBER ground-truth completeness to quantify systematic inflation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline real-data result—16 ptMAGs with >70% completeness and >90% purity (§3.2)—rests on quality scores computed inside ChloroScan. In §2.1, the manually curated plastid marker database is used by binny to guide clustering, and the same database is then used in the summary/QC module to count marker genes, compute marker completeness, and flag contaminant contigs ('without ORFs matching our plastid marker gene database'). Because contigs are drawn into bins precisely because they carry these markers, a bin assembled around marker-bearing contigs can pass a marker-based completeness threshold without covering most of the plastid chromosome; purity filtering can likewise remove genuine plastid contigs that lack the curated markers, inflating purity. This circularity does not invalidate the simulated benchmark (AMBER uses ground-truth mapping, §2.2), so the comparison with MetaBAT2 survives. However, it directly weakens the MIMAG-quality classification of the 16 real bins and the quantitative 70/90 claim. The dinoflagellate minicircle failure is real but explicitly acknowledged (§3.1, Discussion); the marker-QC circularity is more load-bearing because it is unacknowledged and affects every real-data quality assertion.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"ChloroScan is a Snakemake pipeline for recovering plastid genome bins from metagenomic assemblies. It first filters contigs with the deep-learning classifier Corgi, then performs marker-guided binning by adapting binny with a custom plastid marker-gene database, assigns taxonomy with CAT/BAT, and provides QC, annotation, and visualization outputs. The authors benchmark ChloroScan against MetaBAT2 on two CAMISIM-simulated marine metagenomes with known ground truth, reporting higher F1 scores, higher base-level accuracy, and more high-quality plastid bins, including several near-complete single-contig MAGs. They also apply ChloroScan to four Tara Oceans metagenomes and report 16 ptMAGs with completeness >70% and purity >90%, including a bin suggested to represent a novel ochrophyte lineage. The manuscript explicitly acknowledges that dinoflagellate minicircle plastid genomes were not recovered because Corgi's training set lacks such sequences.","tokens_in":16697,"tokens_out":5192,"duration_ms":50817,"significance":"If the results hold, ChloroScan would be a valuable, practical addition to the small set of tools targeting plastid genomes from metagenomes. The simulated benchmark is externally grounded: CAMISIM-generated reads with known genomes and AMBER assessment avoid circularity in the MetaBAT2 comparison. The code, reproducibility scripts, and intermediate data are made available, which is a clear strength. However, the real-data quality claims are weakened by the circular use of the same custom marker database both to guide binning and to compute marker-based completeness/purity, and the headline count of 16 ptMAGs appears to include a bin later identified as a likely contaminant. These are fixable with additional validation and recalculation, so the manuscript is a candidate for major revision rather than rejection.","major_comments":[{"comment":"The real-data quality metrics are circular. The custom plastid marker database is used by binny to guide clustering, and the same database is then used in the summary/QC module to count marker genes, compute marker completeness, and flag contaminant contigs described as 'without ORFs matching our plastid marker gene database'. Bins assembled around marker-bearing contigs can therefore pass marker-based completeness thresholds by construction, while genuine plastid contigs from divergent lineages lacking these markers may be removed, inflating purity. This does not affect the simulated benchmark, which uses AMBER ground truth, but it directly undermines the claimed >70% completeness and >90% purity for the 16 real ptMAGs and their MIMAG classification. Please validate the real bins with an independent approach—for example, split-marker or leave-one-out cross-validation, mapping/coverage e","section":"§2.1, §3.2"}],"minor_comments":[{"comment":"The caption states 'six bins from the sample SAMEA2732360', but the text describes eight bins (Bin 0 through Bin 7) for that sample. Please harmonize the caption and text.","section":"Figure 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely within scope for q-bio.GN. The central simulation-based comparison with MetaBAT2 is sound, but the real-data claims need independent QC validation and a corrected count if the contaminant bin is excluded. The data availability section also contains a version inconsistency: it names 'most updated release v0.1.5' but links to release_v0.1.3."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"ChloroScan is worth engaging with. It fills a real gap—automated plastid genome binning from metagenomes—and it ships with a usable Snakemake workflow plus a custom plastid marker database. The strongest part is the simulated benchmark: CAMISIM ground truth is external, and ChloroScan beats MetaBAT2 clearly on F1 and base-level accuracy for recovering plastid bins. The coverage analysis (roughly a 5x threshold for recovery) is honest and useful.\n\nThe real-data results are softer than they look. The headline 16 ptMAGs with >70% completeness and >90% purity are scored using the same curated marker database that guided binning. Contigs are pulled into bins because they carry those markers, so bins will tend to look marker-complete even if they miss substantial parts of the plastid chromosome. The purity filter—dropping contigs without plastid marker genes—can also expel legitimate plastid contigs that are diverged or fragmented. That circularity specifically inflates the MIMAG-style quality labels. A split-marker validation (half the markers for binning, half for QC) would make the real-data claim credible.\n\nThe dinoflagellate minicircle failure is real but explicitly acknowledged and tied to Corgi's training data; that's an acceptable limitation. Missing a direct comparison with plastiC is a pity, though using its underlying binner MetaBAT2 is a reasonable proxy. The paper is clearly written, code and data are available, and the authors report the multimeric chimera bin rather than hiding it. The marker-QC circularity, however, should be fixed or reframed before the tool is adopted broadly.\n\nI would send this to peer review. The tool is useful and the simulation benchmark is solid. Referees should push for threshold sensitivity analysis and an independent check of the real bins—e.g., read-depth homogeneity or a split-marker design.","headline":"Useful pipeline with a solid simulation benchmark, but the real-data quality numbers hinge on the same marker database used for binning and QC, so they likely overstate completeness and purity.","tokens_in":583,"tokens_out":658,"would_cite":true,"duration_ms":30497,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"ChloroScan recovers plastid genomes from metagenomes by combining deep-learning contig filtering with marker-gene-guided binning, outperforming a generic binner on simulations and recovering 16 plastid genomes from four ocean metagenomes.","keywords":["plastid genomes","metagenome-assembled genomes","metagenomic binning","deep learning contig classification","marker gene database","protist genomics","marine metagenomes","plastid MAGs"],"falsifier":"Simulate a metagenome containing only high-coverage plastid genomes from lineages absent from the classifier's training data, for example dinoflagellate minicircle chromosomes, run ChloroScan, and check whether plastid bins emerge; if the genomes are in the assembly but no bin is recovered, the pipeline's coverage assumption is breached. A second check is to re-run the benchmark on the same simulated communities with several random seeds and see whether the reported F1 advantage over the generic binner persists.","tokens_in":16271,"feed_emoji":"🧬","tokens_out":8531,"duration_ms":71129,"temperature":0.7,"pith_summary":"ChloroScan is an automated pipeline for pulling plastid genomes out of metagenomes, where nuclear protist genomes are often too complex to assemble but organellar genomes are small, high-copy, and phylogenetically informative. The paper argues that a deep-learning contig classifier followed by binning guided by a curated plastid marker-gene database recovers plastid metagenome-assembled genomes more completely and purely than a generic binner on simulated data. Applied to four ocean metagenomes, ChloroScan recovered 16 plastid metagenome-assembled genomes at 70 percent completeness and 90 percent purity, including a likely deep-branching marine ochrophyte lineage with no close sequenced relatives. If correct, this makes plastid genomes accessible from existing metagenomic libraries at scale, easing a bottleneck in discovering protist diversity.","feed_headline":"ChloroScan recovers 16 plastid genomes from four ocean metagenomes","feed_subtitle":"Recovering algal organelle genomes from complex microbial data, including a likely new ochrophyte lineage.","key_machinery":"The load-bearing mechanism is a manually curated database of plastid-encoded marker genes, formatted in the same style as prokaryotic quality-assessment marker sets, which steers the binning algorithm to cluster contigs into complete, pure plastid bins. Around this core, a deep-learning contig classifier pre-filters assemblies to a plastid-enriched set, and contig-level taxonomy is assigned by comparing predicted proteins against a combined reference protein database; a final round of homology searches on the rbcL marker provides fine-grained identification. The marker database and the classifier thresholds are the two settings that most directly determine which plastid lineages are recovere","core_discovery":"The paper's central claim is that plastid genome recovery from metagenomes improves when binning is guided by a plastid-specific marker-gene database rather than by generic prokaryotic markers. ChloroScan's pipeline first uses a deep-learning contig classifier to enrich for plastid contigs, then clusters those contigs with a marker-gene-guided binner, then adds taxonomic assignment and gene prediction. On two simulated marine metagenomes, it recovered more high-quality plastid bins than the benchmark binner, including near-complete single-contig genomes; its F1 and base-level accuracy were higher in both samples. On four real ocean metagenomes it produced 16 plastid metagenome-assembled geno","pith_inferences":["An implication the paper leaves implicit is that the success of marker-gene-guided binning for plastids could generalize to other organellar or extrachromosomal elements, such as mitochondria or plasmids, wherever conserved single-copy genes exist.","If ChloroScan scales to the full set of public metagenomes, it could substantially expand the sampled plastid tree of life; the 85 percent rbcL hit in this paper hints at how much novelty remains in under-sequenced marine lineages.","The systematic failure on dinoflagellate minicircle plastids implies that lineage-specific retraining of the classifier, not just more reference genomes, is needed; users should expect recovery to be biased toward lineages already represented in the training data.","A practical test would be applying ChloroScan to mock communities with known plastid abundances to quantify how the roughly 5x coverage threshold shifts with contig length cutoffs and binning parameters."],"forward_implications":["Existing metagenomic libraries can be re-mined automatically for plastid genomes, without manual or human-guided binning.","Recovered plastid genomes carry phylogenetic markers such as rbcL and coding sequences, so they can feed directly into algal phylogenomics and species-delimitation studies.","Adapting the marker-gene database to mitochondrial genes would extend the same workflow to heterotrophic protists, which lack plastids.","As reference plastid genomes grow, both the classifier and marker database will cover more lineages, improving recovery and taxonomic resolution.","Low-coverage plastid genomes (roughly below 5x average depth) are likely to be missed, so the method's gains concentrate on abundant or deeply sequenced taxa."],"fun_headline_variants":["AI-guided binning finds 16 plastid genomes in ocean data","ChloroScan digs out algal plastid genomes from metagenomes","New tool recovers more plastid genomes from complex microbial data","Deep learning pipeline finds 16 plastid genomes, one new lineage"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The pipeline assumes the deep-learning classifier's training set and the curated plastid marker-gene database cover the target lineage; unusual plastid architectures absent from those resources, such as dinoflagellate minicircles, are filtered out before binning and cannot be recovered.","fun_headline_variants_meta":{"raw":{"variants":["AI-guided binning finds 16 plastid genomes in ocean data","ChloroScan digs out algal plastid genomes from metagenomes","New tool recovers more plastid genomes from complex microbial data","Deep learning pipeline finds 16 plastid genomes, one new lineage"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000165,"raw_usage":{"total_tokens":1087,"prompt_tokens":745,"completion_tokens":342,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":489,"completion_tokens_details":{"reasoning_tokens":280}},"tokens_in":489,"tokens_out":342,"duration_ms":3528,"temperature":1.0,"reasoning_tokens":280,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T10:12:04.029531+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate a metagenome containing only high-coverage plastid genomes from lineages absent from the classifier's training data, for example dinoflagellate minicircle chromosomes, run ChloroScan, and check whether plastid bins emerge; if the genomes are in the assembly but no bin is recovered, the pipeline's coverage assumption is breached. A second check is to re-run the benchmark on the same simulated communities with several random seeds and see whether the reported F1 advantage over the generic binner persists.","supporting_citations":[],"review_version":1}