Why BGPT?
logo

Review Papers Claim by Claim

Check each claim against its experiments, reported results, scope, and stated limitations.Know what the science actually supports before you trust the answer.

Press Enter ↡ to start review



    Explore by Goal












     Quick Explanation


    Paper Review: YaTCM (Yet another TCM Database for Drug Discovery)

    YaTCM is a free web-based TCM systems-style knowledgebase + analysis toolkit that integrates herbs/ingredients with targets, pathways, and diseases, and provides structure-based retrieval, MV-SEA ligand-to-target prediction, and interactive herb–herb/prescription network views. Key value: it packages multi-source curation into one interface and supports multi-layer network exploration for candidate target/pathway hypotheses. Primary skepticism: most β€œmechanism” links are computationally inferred (e.g., similarity and MV-SEA) and therefore require careful validation before any biological conclusions.




     Long Explanation


    YaTCM β€” Yet another Traditional Chinese Medicine Database for Drug Discovery

    DOI: 10.1016/j.csbj.2018.11.002  |  Journal: Computational and Structural Biotechnology Journal  |  Received/Accepted: 2018-08-08 / 2018-11-06

    1) What YaTCM claims to do (from the paper text)

    • Integrated knowledge model: recipes ⇄ herbs ⇄ ingredient molecular structures ⇄ targets ⇄ KEGG pathways ⇄ diseases, stored in a backend database and navigable via UI.
    • Structure retrieval using a JavaScript sketcher to query by structure and perform similarity/substructure search against stored compound structures.
    • Target prediction through MV-SEA (multi-voting similarity ensemble approach) mapping compounds to protein targets via chemical similarity ensembles.
    • Pathway analysis & network pharmacology views, including GO enrichment to rank mapped KEGG pathways and interactive herb–herb networks based on shared targets/diseases/pathways.
    Scientific caveat (explicitly from the modeling style):
    YaTCM’s mechanism links largely come from integration + similarity-based mappings and predictions rather than direct experimental binding/assays, so the outputs are best treated as hypothesis-generating evidence unless validated externally.

    2) Database scale (YaTCM vs other TCM DBs reported by the paper)

    Counts in the bar chart are taken directly from the paper’s comparison table (YaTCM row).

    This chart is limited to categories that appear in the same comparison table columns across systems; the paper reports β€œpredicted targets” for YaTCM but not for the other DBs in that row.

    3) Data pipeline & computational components (what the methods imply)

    3.1 Integrated data sources & curation strategy

    • YaTCM manually integrates multiple TCM-related resources plus biomedical databases and mined literature, including TCMID, Database@Taiwan, TCMSP, TTD, ChEMBL, KEGG, OMIM, PubMed text-mined materials, and books.
    • Deduplication/normalization is described (deduplicate herbs by Chinese/English/Latin name; deduplicate compounds by English/chemical name/structure) and mapping relationships between YaTCM structures and ChEMBL bioactivities and YaTCM entries to KEGG via similarity is constructed.
    • Update cadence: β€œbi-weekly” updates are described via repeated literature text-mining and manual curation workflow.

    3.2 Fingerprints, similarity, and MV-SEA

    YaTCM computes molecular representations and similarity relationships, then uses an ensemble target-prediction approach. It explicitly states that MV-SEA uses four fingerprint types (MACCS, Morgan, Topological, Atom pair) and uses similarity ensemble logic to map compounds to targets in ChEMBL.

    Skeptical implication: similarity-to-target approaches can be sensitive to dataset coverage, fingerprint choice, and the assumption that β€œchemical similarity implies target/pathway similarity.” The paper does not provide in-text target-prediction ROC/AUPR metrics in the provided excerpt, so model reliability is not quantified here.

    3.3 Pathway mapping and enrichment

    • YaTCM maps compounds to KEGG human metabolic pathways by calculating structural similarity between TCM compounds and KEGG compounds/metabolites; it also maps therapeutic targets to KEGG human signal transduction pathways.
    • To rank the importance of mapped pathways, the paper applies gene ontology enrichment analysis and uses p-value/q-value style statistical indicators to rank pathway hits for the Yu Ping Feng case study.
    Important limitation (statistical interpretability):
    When enrichment is driven by inferred targets/pathways, enrichment p-values quantify consistency with the curated mapping pipeline, not necessarily evidence of direct biochemical causality. This is not a flaw in enrichment itself, but it shifts what conclusions are warranted.

    4) Case study: Yu Ping Feng (reported network layers)

    YaTCM demonstrates a prescription-level case study for Yu Ping Feng using three herbs: Astragalus membranaceus (AM), Atractylodes macrocephala (RA), and Saposhnikovia divaricata (SD). The paper reports the case study network sizes: total 497 molecules, 379 pathways, 5,502 targets, and 1,102 diseases.

    The paper also reports that pathway enrichment is split into β€œranked by statistical significance” vs β€œnot ranked due to low enrichment information,” and it provides an example of a pathway layer network that connects prescription–herb–molecule–target–pathway–disease.

    Example mechanistic narrative in the paper (inferred):
    The paper highlights TGF-Ξ²1 and ERK/MAPK signaling as part of its mechanistic mapping narrative for pulmonary fibrosis and ties these to Baicalin’s reported downregulation of a pathway involving TGF-Ξ²1-induced ERK signaling.

    5) Critical review: what’s strong, what’s missing, and what could mislead

    5.1 Strengths (evidence grounded in the paper text)
    • End-to-end integration: the paper explicitly describes mapping across multiple entity types (prescription/herb/ingredient/target/pathway/disease) and supporting interactive navigation.
    • Explicit computational modules: structure retrieval, MV-SEA, pathway mapping, network analysis are all described with specific ingredients (e.g., fingerprints and the existence of ADMET filtering and a web stack).
    5.2 Main limitations & blind spots (scientific)
    • Prediction uncertainty not quantified in the provided excerpt: MV-SEA is described, but (in the excerpt provided to me) there are no explicit predictive performance metrics (e.g., AUPR/ROC, calibration, top-k accuracy) attached to YaTCM outputs. That makes it hard to judge error rates or false-positive rates for β€œpotential targets.”
    • Inference stacking: pathway enrichment uses gene lists/targets derived from similarity mappings/predictions; this can inflate apparent pathway coherence without demonstrating direct interaction or pathway activation causality.
    • Coverage bias across TCM ingredients: integration depends on what exists in source databases and what can be mined from books/PubMed. That typically yields uneven coverage (well-studied herbs/ingredients are more represented). YaTCM describes manual curation and deduplication, but the excerpt doesn’t quantify completeness per herb or confidence tiers for each edge.
    • External validation scope: the case study uses external experimental literature to claim consistency, but the paper’s mechanistic network still needs prospective experimental testing of high-ranked targets/pathway mechanisms in relevant biological contexts.
    5.3 What would most disprove the paper’s practical utility?
    • Demonstrate that MV-SEA predicted targets (top-ranked) show poor enrichment for known TCM–target interactions relative to baselines, indicating low signal-to-noise for predictions.
    • Show that pathway enrichment computed from YaTCM-mapped sets does not outperform random mapping or structural-neighborhood baselines for retrieving disease-relevant pathways.

    6) Practical β€œhow to use” suggestions (hypothesis-first workflow)

    1. Use structure similarity retrieval to extract candidate compounds from YaTCM given a substructure/feature hypothesis.
    2. Run MV-SEA target prediction for those compounds; interpret β€œpredicted targets” as candidate proteins for experimental prioritization, not as confirmed binders.
    3. Use pathway + GO enrichment to identify pathway convergence among predicted targets and curated targets; treat p-values/q-values as pipeline statistics that guide prioritization.
    4. Use herb–herb / prescription networks to find shared targets/diseases across herbs; this helps identify β€œcommon mechanistic hypotheses” but should still be experimentally tested.


    Feedback:    

    Updated: April 30, 2026

     BGPT Paper Review


    Study Novelty

    80%

    Novelty is primarily in the packaging of multi-entity TCM structural integration (herbs/ingredients/targets/pathways/diseases) plus a unified MV-SEA + pathway/network toolkit interface and interactive herb–herb/prescription exploration, rather than in introducing a wholly new prediction paradigm. The paper positions YaTCM as integrating large-scale structural data with multiple analysis tools.


    Scientific Quality

    80%

    Scientific quality is supported by (i) explicit module descriptions, (ii) stated integration and deduplication steps, (iii) clear use of established knowledge resources (KEGG/ChEMBL/TTD/OMIM cited in the text), and (iv) a reproducible workflow description (fingerprints, enrichment, mapping logic) in the methods excerpt. Main quality concern: predictive/model performance is not shown in the provided excerpt, and interpretability/uncertainty quantification for predicted edges is not evidenced here.


    Study Generality

    70%

    Generality is moderate-high for systems pharmacology workflows because it generalizes across herbs/prescriptions/targets/pathways/diseases and provides reusable computational interfaces. It is less general than a fully agnostic interaction assay because it is constrained by the coverage and curation scope of the integrated resources and by similarity-based inference.


    Study Usefulness

    90%

    Practical usefulness is high as a hypothesis-generation platform: structure query β†’ candidate compounds β†’ predicted targets β†’ pathway enrichment and interactive networks for herb/prescription convergence. This is directly aligned with common early-stage target/pathway discovery workflows.


    Study Reproducibility

    70%

    The paper describes workflow steps (data integration, similarity-based mapping, fingerprint ensemble types, enrichment usage) and the web stack, which supports partial reproducibility. Full reproducibility is limited by the availability of exact code/data and by the lack of explicit predictive evaluation metrics in the provided excerpt.


    Explanatory Depth

    70%

    Explanatory depth is decent for a database paper: it explains the data model, mapping logic, and how users can traverse layers of inference. It is less mechanistically deep in a biochemical causality sense because target/pathway connections are often computationally inferred.


    🎁 Authors: Collect 395 Free Science Tokens (β‰ˆ $39.5 USD)

    Claim My Author Tokens

    Use for 98 days of free BGPT access (4 tokens = 1 day) or trade/sell (β‰ˆ $39.5 USD)

     Top Data Sources ExportMCP


     Analysis Wizard



    It will parse the paper’s reported YaTCM and case-study counts, generate comparative charts, and compute proportional contributions per herb to molecules/targets/pathways/diseases.



     Hypothesis Graveyard


    The case study’s pulmonary fibrosis narrative may be over-attributed to ERK/TGF-Ξ²1 solely due to literature popularity and network hub effects, rather than being a uniquely discriminative mechanism arising from the YaTCM mapping pipeline.


    Because the system is similarity-driven, high agreement between YaTCM pathways and known literature may reflect database overlap (same widely studied targets/pathways) rather than a new mechanistic signal beyond what curated resources already contain.

     Science Art


    Paper Review: YaTCM: Yet another Traditional Chinese Medicine Database for Drug Discovery Science Art

     Science Movie


    Make a narrated HD Science movie for this answer ($32 per minute)




     Discussion


    Stay current without chasing every paper.

    Know what changed, what holds up, and what remains uncertain. Every Friday. No ads.


    My BGPT






     Trending