The Evidence Layer for Scientific AI

 Long Analysis Plan Copied

What Orthogonal Validation Means Here

The strongest evidence for validating cyanopeptide class annotations comes from the 2026 Muskegon/Lake Erie study, which stacks four independent lines: HRMS accuracy, MS/MS diagnostic product ions, full 1D/2D NMR structure elucidation, and metagenomic detection of biosynthetic genes (ociB for cyanopeptolin/micropeptin pathways) in the same bloom material MassQL/NMR/metagenomic cyanopeptide validation study. Each layer corrects a different failure mode: HRMS rules out formula errors, NMR resolves connectivity and stereochemistry that MS/MS cannot, and biosynthetic genes independently corroborate the chemical class call — a triangulation the Swiss multi-class quantification study implicitly supports by requiring matrix-matched calibration across five classes Swiss lake multi-class cyanopeptide quantification.

Observed data: all isolated compounds fall within typical ±5 ppm HRMS windows, with micropeptin 1010 at 0.39 ppm and the weaker 2.83 ppm value for micropeptin 966 flagged as the least precise annotation HRMS error table for isolated compounds.

Where the Annotation Chain Is Weakest

Class-level ≠ structure-level. The MassQL product-ion queries (m/z 404, 243, 134 for micropeptin/cyanopeptolin) are sensitive but not isomer-resolving, and the authors themselves concede in-source fragmentation and adduct formation can inflate apparent chemical diversity Stated limitations of MassQL annotation. Putative new micropeptins 1038 and 1024 rest only on network nodes and hexanoic-acid ion searches — no NMR — so their annotations should be treated as level-2 (probable structure), not level-1 (BGPT inference). Quantification gap: the discovery workflow is semi-quantitative at best; the Swiss study shows ng/L quantification requires isotope-labeled or matrix-matched standards, with recoveries only ~30–70% Recovery and internal-standard gaps in quantification. Genomic corroboration is sparse: only one gene (ociB) was linked at one site; broader Microcystis work shows most BGCs remain orphan, so gene-to-metabolite inference is the exception, not the rule Microcystis BGC–metabolite linkage review.

Battle-Testing: Controls, Bias, and Falsification

Three blindspots stand out. First, sample provenance: the fraction library was Biosortia-supplied (founder/CEO reviewed the manuscript), a mild financial-incentive bias acknowledged in the disclosure Conflict-of-interest disclosure. Second, temporal bias: short pre-bloom-maximum sampling windows can systematically miss microcystins and bias class prevalence estimates (BGPT inference from stated limitation). Third, the absence of microcystins in Lake Erie field samples (not detected in this dataset) is a negative result that contradicts regional literature expectations — it should be cross-checked with a targeted quantitative method like the Swiss online-SPE workflow before being treated as real Field negative result for microcystins. Falsification paths are well-defined: independent replication failing to find micropeptin 1010/966 in Muskegon fractions, MassQL queries failing on external datasets, or NMR-inconsistent re-isolation would overturn the claims Pre-registered falsification criteria.

Verdict and What Would Change It

Conclusion (high confidence for isolated compounds; moderate for class-level workflow): the four-layer design — HRMS ± NMR ± diagnostic MS/MS ± biosynthetic gene — is currently the best-practice template for orthogonal validation, with NMR-confirmed compounds at the strongest evidence tier and MassQL-only network nodes at the weakest (BGPT inference from reported methods). The conclusion would change if: (a) isomer-resolving methods (e.g., ion mobility or re-isolation) showed MassQL nodes routinely confound congeners; (b) matrix-matched quantification of the same Lake Erie samples detected microcystins that DDA-MS/MS missed; or (c) independent datasets failed to reproduce the class-level workflow. Related method advances — CZE-MS/MS screening 14 cyclic peptides in under 7.5 min Rapid CZE-MS/MS cyanopeptide screening and biodegradation-driven LC-MS/MS confirmations in Baltic consortia Baltic biodegradation LC-MS/MS evidence — offer complementary orthogonal checks (migration-time independence and biological turnover, respectively).