Targeted methods measure the constituents you expect, non-targeted screening looks for the ones you did not

Liquid chromatography (LC) separates the constituents of an extract by retention time; electrospray ionization (ESI) charges them; a high-resolution mass spectrometer (HRMS, such as an Orbitrap or a quadrupole time-of-flight) records their mass-to-charge ratio (m/z) to within a few parts per million and fragments selected ions (MS/MS).

Targeted analysis measures a short list of named analytes against reference standards and is validated for its purpose. Suspect screening searches the data for compounds on a prepared list (formula, m/z, sometimes MS/MS) without a standard for each. Non-targeted screening treats every detected signal as a possible compound. The NORMAN guidance notes that screening allows detection "without upfront selection and purchase of standards" and that target methods "are typically optimised for a small set of compounds".

Hollender J, Schymanski EL, Ahrens L, et al. NORMAN guidance on suspect and non-target screening in environmental monitoring. Environ Sci Eur. 2023;35:75. ICH Q2(R2), adopted November 1, 2023.

ICH Q2(R2) covers procedures for identity, assay, purity and impurity testing and sets out the performance characteristics to demonstrate (specificity or selectivity, working range, accuracy and precision, and detection or quantitation limits where relevant); the method-development page shows them by test type. Non-targeted screening has no equivalent harmonized validation guideline; the NORMAN guidance sets out quality assurance and states that "blanks are a requirement for NTS", because features present in blanks must be removed.

The symbols are EUTECNIQ's summary of these documents, not a ranking from any one source. Absolute quantitation without a reference standard is possible only as an estimate; NORMAN treats quantification without standards as a separate topic.

ICH Q2(R2) Validation of Analytical Procedures (2023); see analytical method development. Hollender et al. 2023 (NORMAN). Schymanski EL, et al. Environ Sci Technol. 2014;48(4):2097–2098.

Kind and Fiehn showed that "mass accuracy is insufficient even at less than 1 ppm" to give a unique formula in higher mass ranges, and proposed seven heuristic rules to remove chemically impossible or improbable formulas. Of their rules, isotope-pattern scoring was the most efficient filter; it needs the measured abundances of the M+1 and M+2 isotopologues, which this teaching tool does not use.

Kind T, Fiehn O. Metabolomic database annotations via query of elemental compositions: mass accuracy is insufficient even at less than 1 ppm. BMC Bioinformatics. 2006;7:234. Kind T, Fiehn O. Seven Golden Rules for heuristic filtering of molecular formulas obtained by accurate mass spectrometry. BMC Bioinformatics. 2007;8:105.

A feature is a detected signal defined by its m/z, retention time and intensity in each sample. Feature finding, alignment across samples and gap filling are implemented, for example, in MZmine 3, XCMS (nonlinear retention-time alignment, matched filtration, peak detection and matching) and MS-DIAL (deconvolution of data-independent MS/MS). Feature-based molecular networking (FBMN) in GNPS takes such a feature table and its MS/MS spectra, which allows quantitative comparison and the separation of isomers.

Schmid R, et al. Integrative analysis of multimodal mass spectrometry data in MZmine 3. Nat Biotechnol. 2023;41(4):447–449. Smith CA, et al. XCMS. Anal Chem. 2006;78(3):779–787. Tsugawa H, et al. MS-DIAL. Nat Methods. 2015;12(6):523–526. Nothias LF, et al. Feature-based molecular networking in the GNPS analysis environment. Nat Methods. 2020;17(9):905–908.

Molecular networking (Watrous et al. 2012; Wang et al. 2016) compares every MS/MS spectrum with every other. The modified cosine score counts a fragment pair as matched if the two m/z values agree, or if they agree after adding the difference between the two precursor masses, so a methylation or an extra hydroxyl does not break the match. Spectra above a score threshold are joined by an edge; clusters of related structures emerge.

For this page EUTECNIQ downloaded the 802 spectra of the LEAFBOT library and wrote its own implementation, using the defaults in the GNPS documentation: peaks within ±17 Da of the precursor removed; only the 6 most intense peaks kept in any ±50 Da window; spectra with fewer than 5 fragments left unconnected; an edge where cosine ≥ 0.7 with at least 6 matched fragments; each node keeps at most its 10 best neighbors (mutual); clusters are capped at 100 nodes; fragment tolerance 0.02 Da, the documented value for high-resolution data. Intensities were not transformed. Only [M+H]⁺ (positive) or [M−H]⁻ (negative) spectra were used.

Edges reflect shared fragments, not shared biosynthesis. In positive mode, steviol glycosides and flavonoid glycosides fall in one cluster, connected through sugar-related fragments; in negative mode they separate. Library names were normalized for display (lower case, obvious misspellings corrected); the edge list keeps the original names.

Watrous J, et al. Mass spectral molecular networking of living microbial colonies. Proc Natl Acad Sci USA. 2012;109(26):E1743–E1752. Wang M, et al. Sharing and community curation of mass spectrometry data with Global Natural Products Social Molecular Networking. Nat Biotechnol. 2016;34(8):828–837. GNPS documentation, molecular networking parameters (accessed October 24, 2026). LEAFBOT library page, GNPS (accessed October 24, 2026).

The five levels separate what was measured from what was inferred. A level-4 formula comes from exact mass with isotope or adduct evidence; a level-3 annotation names a candidate structure, substituent or class; level 2 needs a library spectrum match (2a) or diagnostic evidence (2b); level 1 requires a reference standard measured with MS, MS/MS and retention time. The example m/z 465.1028 is the [M+H]⁺ ion of C₂₁H₂₀O₁₂ (for example hyperoside, a quercetin hexoside), computed from NIST masses.

Schymanski EL, Jeon J, Gulde R, Fenner K, Ruff M, Singer HP, Hollender J. Identifying small molecules via high resolution mass spectrometry: communicating confidence. Environ Sci Technol. 2014;48(4):2097–2098. Hollender et al. 2023, Table 7.

Dereplication means recognizing known compounds early, so that effort goes to the unknowns. Structure databases answer "is this structure known, and from which organism?"; spectral libraries answer "does this spectrum match a measured reference?"; in silico tools rank candidates when no reference spectrum exists.

• COCONUT 2.0: an open collection of natural products from 63 data collections; 695,133 unique structures in the September 2024 release. Stereoisomers are kept as separate molecules under one identifier with a suffix. Data released under CC0. The October 2026 dump used here has 738,794 entries and 479,690 unique 2D structures.
• LOTUS: an open resource of more than 750,000 referenced structure–organism pairs.
• GNPS spectral libraries: community reference MS/MS spectra; spectra contributed directly to GNPS carry a CC0 license by default.
• SIRIUS 4 with CSI:FingerID searches structure databases from MS/MS when no library spectrum exists; CANOPUS predicts 2,497 compound classes from fragmentation spectra; NPClassifier assigns structures to 7 pathways, 70 superclasses and 672 classes.

Six LEAFBOT [M+H]⁺ spectra had fewer than 5 fragments after filtering and were not networked. Five of the six spectra without a COCONUT match are of three aminomethylbenzothiazoles, synthetic reference compounds; the sixth is rebaudioside M, a steviol glycoside whose structure had no connectivity match in the October 2026 dump.

Chandrasekhar V, et al. COCONUT 2.0: a comprehensive overhaul and curation of the collection of open natural products database. Nucleic Acids Res. 2025;53(D1):D634–D643. COCONUT download page (October 2026 release; CC0). Rutz A, et al. The LOTUS initiative for open knowledge management in natural products research. eLife. 2022;11:e70780. GNPS documentation, library license. Dührkop K, et al. SIRIUS 4. Nat Methods. 2019;16(4):299–302. Dührkop K, et al. CANOPUS. Nat Biotechnol. 2021;39(4):462–471. Kim HW, et al. NPClassifier. J Nat Prod. 2021;84(11):2795–2807.

Dewick organizes secondary metabolism by building block: the acetate pathway (fatty acids and polyketides), the shikimate pathway (aromatic amino acids and phenylpropanoids, including coumarins, flavonoids and stilbenes), the mevalonate and methylerythritol phosphate (MEP) pathways (terpenoids and steroids), alkaloids grouped by their amino-acid origin, peptides and carbohydrates. NPClassifier uses seven matching pathways. Counts here use one record per InChIKey connectivity block (stereoisomers counted once), so they are lower than the database's entry count of 738,794.

Dewick PM. Medicinal Natural Products: A Biosynthetic Approach. 3rd ed. Chichester: Wiley; 2009 (ISBN 978-0-470-74168-9). Kim HW, et al. J Nat Prod. 2021;84(11):2795–2807. COCONUT October 2026 dump.

Each model is one low-energy conformer computed by EUTECNIQ with RDKit (ETKDG version 3 embedding, then MMFF94 optimization) from the PubChem SMILES; the molecular formula was checked against PubChem. It is not a crystal structure and not the only shape the molecule adopts. Drag or use the arrow keys to rotate; dashed bonds are aromatic.

Ginkgolides A, B, C and J are described as diterpene trilactones and bilobalide as a sesquiterpene trilactone (EFSA FEEDAP Panel 2024); the NPClassifier label in the dump is therefore wrong for ginkgolide B.

PubChem compound records (CIDs above; checked October 24, 2026). Wang S, Witek J, Landrum GA, Riniker S. J Chem Inf Model. 2020;60(4):2044–2058 (ETKDG v3). Halgren TA. J Comput Chem. 1996;17(5–6):490–519 (MMFF94). RDKit 2026.09.1. EFSA FEEDAP Panel. EFSA J. 2024;22(4):e8733. Strømgaard K, Nakanishi K. Angew Chem Int Ed. 2004;43(13):1640–1658.

A Bemis–Murcko scaffold keeps the ring systems and the linkers between them and removes side chains; RDKit keeps atoms double-bonded to the scaffold, so carbonyl groups of coumarins, flavones and xanthones remain. Stereochemistry was removed first, so stereoisomers share a scaffold. 55,326 structures (mostly fatty acids) have no ring and therefore no scaffold. Steroid ring systems appear more than once because RDKit scaffolds keep ring double bonds.

Bemis GW, Murcko MA. The properties of known drugs. 1. Molecular frameworks. J Med Chem. 1996;39(15):2887–2893. RDKit 2026.09.1 (MurckoScaffold). COCONUT October 2026 dump.

Descriptors were computed with RDKit for each sampled structure: molecular weight, Crippen logP, topological polar surface area (TPSA), hydrogen-bond donors and acceptors, rotatable bonds, Fsp3, ring and aromatic-ring counts, oxygen and nitrogen counts and stereocenters. PC1 rises with size and polarity (molecular weight, TPSA, donors, acceptors, oxygen count); PC2 rises with flexibility and saturation (rotatable bonds, Fsp3, logP) and falls with aromatic rings. Natural products have long been noted to differ from synthetic compounds in stereocenters, aromatic rings, saturation and heteroatom ratios (Feher and Schmidt 2003). Sampling one record per connectivity block avoids weighting molecules with many recorded stereoisomers.

Feher M, Schmidt JM. Property distributions: differences between drugs, natural products, and molecules from combinatorial chemistry. J Chem Inf Comput Sci. 2003;43(1):218–227. RDKit 2026.09.1. COCONUT October 2026 dump (CC0).

In practice, release testing of a botanical ingredient under 21 CFR Part 111 relies on identity, strength and contaminant methods that are scientifically valid for the matrix. Non-targeted screening would support raw-material comparison, adulteration investigations and discovery, and its findings would be reported with a confidence level. Until EUTECNIQ has its own laboratory, such work would be placed with qualified partner laboratories.

Related pages: analytical methods by question, method development and validation, discovery and characterization.