Synotis solidaginea Multifunctional Mass Spectrometry Database

There are 31 files each in positive and negative ion modes, i.e., four groups CXGF/CXMR/CXSR/HYXJ with 6 biological replicates each (24 in total) + 3 QC + 4 SQC pooled samples. The meanings of the four group codes need to be supplemented by the data provider; this dataset retains the original codes and uses them in between-group comparisons.Structural databases: PubChem and ChEMBL; SMILES and structural identifiers were obtained by InChIKey.Activity database: ChEMBL (release 37) REST API; activity, mechanism, molecular properties, and indications.Natural product annotation: LOTUS, NPClassifier, and ClassyFire; 753 compounds were cross-annotated.Computational prediction: ADMET-AI 2.0.1 and RDKit 2026.03.6; ADMET, drug-likeness, and structural alerts.2 Data Structuremaster_table.csv contains 411 columns × 2,206 rows, with InChIKey as the primary key, and can be divided into the following column groups:Compound identifiers: id, name, chinese_name, formula, inchikey, cas, hmdb, kegg, pubchem_cid, chembl_id.Structural layer: smiles, smiles_std, exact_mass, mw_rdkit, heavy_atoms, formula_match.Mass spectrometry reference layer: mz_[M+H]+, mz_[M+Na]+, mz_[M+NH4]+, mz_[M+K]+, mz_[M-H]-, mz_[M+Cl]-, mz_[M+HCOO]-, nl_*, M+1_pct, M+2_pct, db_mz_adduct, db_mz_ppm.Measured detection layer: status, confidence, matched_by, obs_mz, ppm_error, obs_rt_median_s, rt_dev_s, n_groups_detected, pos_detected_ratio, neg_detected_ratio, group_mean_*, cxgf_vs_cxmr_log2FC, mw_p_value, ms2_confirmed, ms2_ppm_error.Activity and pharmacology layer: n_records, n_targets, best_pchembl, best_target, best_type, mechanism_of_action, atc_class, max_phase, indications, black_box_warning, withdrawn_flag.ADMET layer: admet_* (52 endpoints: 31 probability values + *_decision columns, 21 regression variables), *_drugbank_approved_percentile.Drug-likeness layer: qed, np_likeness, sa_score, ro5_violations, ro5_pass, veber_pass, egan_pass, ghose_pass, alert_pains_hit, alert_medchem_nibr_hit.Annotation layer: super_class/class/sub_class, npc_superclass/npc_class/npc_pathway, cf_superclass/cf_class, in_lotus.3 Construction Methods3.1 Structural LayerSMILES were retrieved from PubChem by InChIKey (2,145 hits out of 2,206); for those not found, ChEMBL was used as a fallback.

RDKit was used for standardization (largest fragment, charge removal), and monoisotopic exact mass and molecular formula were calculated and compared element by element with the original molecular formula. Successful parsing: 2,149/2,206; formula validation: 2,144/2,149 consistent. The remaining 57 compounds (mainly phospholipids, lipids, and a few plant components) had stereochemical InChIKeys not included in PubChem or ChEMBL, but their molecular formulas were complete; therefore, exact masses were calculated directly from their molecular formulas using elemental monoisotopic masses and labeled level="formula_only".3.2 Mass Spectrometry Reference LayerFor all 2,206 compounds, theoretical m/z values of seven common adducts were calculated from the exact monoisotopic mass (positive ions [M+H]+, [M+Na]+, [M+NH4]+, [M+K]+; negative ions [M−H]−, [M+Cl]−, [M+HCOO]−), together with product m/z values for five common neutral losses (−H2O, −CO2, −CH3, −C6H10O5, −C6H12O6) and relative isotope intensities for M+1/M+2/M+3 (calculated exactly from the molecular formula, including characteristic Cl/Br isotope ratios).

At the same time, theoretical masses were used to infer the ion form of the mz column in the original data table: 1,899 could be assigned (predominantly [M+H]+ for positive ions and [M−H]− for negative ions), while 307 deviated by >10 ppm from any common adduct and were labeled db_mz_unassigned.3.3 Measured Detection Layer (MS1)For each candidate ion m/z, the maximum intensity within a ±10 ppm window in each MS1 scan was used to construct an extracted ion chromatogram (XIC, x-axis = retention time).

If the peak height exceeded the sample noise threshold, it was considered detected in that sample.Criteria: ±10 ppm mass window; ±30 s retention time window (sensitivity results for ±60 s and unrestricted RT are also reported); noise threshold = max(10 × median intensity of all MS1 peaks in the file, 5,000); each compound required support from at least 3 replicate samples.Matching anchors: both the mz value reported in the library table itself and each adduct derived from the exact mass were used, recorded as matched_by="db_mz" or matched_by="calc_adduct:<adduct name>".Decoy control: after shifting all retention times in the library by +180 s, the identical workflow was rerun to obtain empirical false match rates (10.5% for positive ions, 13.7% for negative ions; based on candidate detection counts).Results: 1,503/2,206 compounds were detected; of these, 1,463 in ≥1 group, 1,316 in ≥2 groups, 1,163 in ≥3 groups, and 812 in all 4 groups; confidence levels were high 968, medium 495, low 40.3.4 Measured Detection Layer (MS2)MS2 (DDA) was acquired only from 8 SQC pooled-sample files, totaling 25,164 MS2 spectra.

The eight injections each covered a precursor ion window (approximately 70–200, 190–400, 390–600, and 590–1200 m/z), together covering the full 70–1200 m/z range. MS2 spectra were merged by precursor m/z into 8,175 consensus spectra (deisotoped and background-subtracted) and compared with candidate ions of library compounds.Tolerances: precursor matching 10 ppm, fragment formula assignment 15 ppm.Confirmation rule: precursor |ppm| ≤ 1.0 and ≥2 spectra, ≥3 consensus peaks, and consistent RT; under this rule, the compound-level decoy false discovery rate (FDR) was 5.6% (decoys were controls with shifted precursor m/z).A complete rule grid (ppm 0.5–5.0 × number of spectra 1–5 × consensus peaks 1–10 combinations) and corresponding FDRs are also provided.Results: 1,079 compounds were confirmed at the MS2 level (ms2_confirmed), with 13,850 matching records in total (including 2,783 decoy rows).3.5 Activity and Pharmacology Layer (ChEMBL Measured Records, Not Predictions)Structural matching in ChEMBL by InChIKey hit 948/2,206 compounds.

Based on this, the following were retrieved: 96,448 activity records (involving 3,675 ChEMBL targets), 46 mechanism-of-action records, 948 molecular property records, and 1,177 indication records.919 compounds had activity records, and 573 had pChEMBL quantitative values (−log10 molar IC50/Ki/EC50, etc.); best_pchembl, best_target, and activity type were given according to the strongest activity.Clinical phases: 51 marketed (max_phase=4), 34 phase III, 47 phase II, and 9 phase I; 64 had ATC classifications.Safety fields: 9 with FDA boxed warnings and 8 withdrawn.Administrative placeholder targets in ChEMBL (e.g., "Unchecked", "NON-PROTEIN TARGET", "ADMET") were excluded during aggregation.3.6 ADMET Layer (Machine Learning Predictions)ADMET-AI 2.0.1 was used to predict 52 endpoints for 2,149 parseable compounds, with no missing values:31 binary endpoints (probability output + 0/1 decision column, threshold 0.5; this version did not provide endpoint-specific thresholds with the package, so the TDC binary classification convention was uniformly applied): absorption (HIA_Hou, Bioavailability_Ma, PAMPA_NCATS, Pgp_Broccatelli), distribution (BBB_Martins), metabolism (CYP1A2/2C19/2C9/2D6/3A4 inhibition, CYP2C9/2D6/3A4 substrate), toxicity (hERG, AMES, DILI, ClinTox, Carcinogens_Lagunin, Skin_Reaction), nuclear receptor (NR-AR, NR-AR-LBD, NR-AhR, NR-Aromatase, NR-ER, NR-ER-LBD, NR-PPAR-gamma), stress response (SR-ARE, SR-ATAD5, SR-HSE, SR-MMP, SR-p53).21 regression endpoints (physicochemical properties and pharmacokinetic parameters): molecular weight, logP, HBD/HBA, TPSA, QED, number of stereocenters, aqueous solubility, lipophilicity, hydration free energy, Caco-2 permeability, plasma protein binding, volume of distribution, half-life, hepatocyte/microsomal clearance, LD50, etc.Each endpoint also includes _drugbank_approved_percentile (percentile in the reference population of marketed drugs; not used for threshold determination).Positive rates for selected endpoints (2,149 compounds): human intestinal absorption 92.2%, oral bioavailability 84.1%, blood-brain barrier penetration 61.2%, DILI hepatotoxicity 38.9%, hERG cardiotoxicity 29.7%, CYP3A4 inhibition 17.7%, and AMES mutagenicity 11.0%.3.7 Drug-Likeness and Structural Alert LayerRDKit and medchem calculations: Lipinski's rule of five (1,378 passed, 64.1%), Veber (1,401), Egan (1,408), Ghose (611), QED (median 0.453; 924 with ≥0.5), NP-likeness (median 1.56), synthetic accessibility SA score, and structural alert catalogs such as PAINS (183 hits, 8.5%), NIBR, BRENK, and ChEMBL rules.Interpretation of alert results should be catalog-specific: the 1,340 NIBR hits (62.4%) appear high, but the NIBR ANNOTATION layer contains functional groups common in natural products (polyphenols, glycosides, Michael acceptors, etc.); for a plant secondary metabolite library, a high hit rate is expected and does not mean these compounds are problematic.

Only the EXCLUDE/FLAG layers and PAINS are closer to "requiring attention". This dataset therefore presents them stratified by severity and does not merge them into a single "alert burden" metric.4 Coverage StatisticsItemCountTotal compounds2,206 (POS 1,231 / NEG 975)Successful structure parsing2,149 (formula validation consistent 2,144/2,149)Measured MS1 detections1,503Measured MS2 confirmations1,079ChEMBL structural matches948With activity records / with pChEMBL919 / 573Marketed drugs (max_phase=4)51ADMET prediction coverage2,149LOTUS / NPC / ClassyFire annotations753 / 688 / 753Major chemical superclass distribution: shikimate-phenylpropanoids 562, terpenoids 486, fatty acids 271, alkaloids 220, polyketides 100, amino acids and peptides 72, carbohydrates 61.5 File ListFileDescriptionindex_database.htmlInteractive database itself (self-contained single file; double-click to open in a browser)master_table.csvMaster table, 411 columns × 2,206 rowsREADME.mdDetailed description: coverage statistics, methodology, evidence semantics, known limitationsdatabase_statistics.jsonCoverage rates, decision thresholds, data quality flagsdetection_summary.jsonMeasured matching parameters, decoy FDR, sensitivity analysis, and rule gridMANIFEST.txtPackage manifest and evidence semantics reminder6 UsageDouble-click index_database.html and open it in a browser; no server or software installation is required.

It supports full-text search (name / Chinese name / molecular formula / CAS / InChIKey / ChEMBL ID), filtering by polarity, chemical superclass, detection status, MS2 confirmation, clinical phase, activity, and drug-likeness, column sorting and pagination. Clicking any row expands the complete details of that compound (structure, adducts and isotopes, measured detection evidence, activity targets, ADMET endpoints, drug-likeness), and the current filtered results can be exported as CSV.For batch analysis or import into other systems, use master_table.csv.

Decision thresholds and parameter sources are given in database_statistics.json and detection_summary.json. Structural images are referenced directly through the PubChem image API using <img>; in an offline environment, structural images will not display, but other functions are unaffected.7 Evidence Levels and Interpretation

Notes

This dataset provides explicit evidence labels for data from different sources. Do not read "no evidence" as "does not exist":status="detected": MS1 evidence was found in the raw data under the established thresholds.status="no_MS1_evidence": no MS1 evidence was found under the established thresholds; this does not mean the compound does not exist.status="formula_only": a provenance label for the structure, indicating that the exact mass was derived from the molecular formula; these compounds were tested through the same workflow.ms2_confirmed=true: spectral-level confirmation was obtained at the SQC measured DDA MS2 level (FDR 5.6%).Empty ChEMBL fields: no database record, not equivalent to no activity.admet_, qed, np_likeness, alert_: computational predictions, for research reference only, and do not constitute conclusions on efficacy or safety.Evidence hierarchy: MS2 spectral confirmation (1,079) > MS1 exact mass + retention time (1,503) > computational prediction only (2,149).8 DisclaimerThe activity and pharmacology data in this dataset are existing database records.

ADMET, drug-likeness, and structural alerts are machine learning predictions or rule-based calculation results. All are for scientific research reference only and do not constitute any conclusions on efficacy, safety, or clinical use. Interpretation of experimental data (mass spectrometry) is limited by the acquisition mode and decision thresholds.