Because the HPLC, diode-array detector, and mass spectrometer, and person-hours, are cost-shared in kind by the UCSB Mammalian Synthetic Biology Foundry, ExFab, and other UCSB facilities, the entire microgrant will fund the materials and methods needed to establish this workflow and capture the first set of molecules. Reference standards set the validation ground truth and chromatography and mass-spec consumables cover the pilot runs.

Additional Information

Approach:

Using the NCI extract library and the automated instruments at the UCSB Mammalian Synthetic Biology Foundry, we will build a high-throughput workflow that pulls each extract apart and reads its components in three steps:

1. Separate. Run one extract through liquid chromatography; the column resolves its molecules in time, eluting them one after another.
2. Read continuously. An inline diode-array detector records a full UV-Vis absorbance spectrum (250-800 nm) up to tens of times per second, capturing a clean spectrum for each compound as it passes.
3. Identify. An inline mass spectrometer records accurate mass and an MS/MS fragmentation fingerprint for each peak. High-confidence identities come from MS library matching (mzCloud, GNPS); the remainder are retained as accurate-mass features at a labeled, lower-confidence tier, identifiable later by NMR if needed.

The NCI library. The Foundry holds one of the few complete copies of the NCI Program for Natural Product Discovery prefractionated library: over 230,000 extract and 326,000 fractions generated from a repository of natural-product extracts assembled since 1986 from more than 25 countries, spanning thousands of plant, marine, fungal, and algal genera, and among the largest and most chemically diverse natural-product collections in existence (dctd.cancer.gov/programs/dtp/organization/npb/npnpd). Equally important, the library sits alongside the automation that ran a 370,830-compound iPSC screen (Wong … and Wilson, et al., Cell 2025).

Anticipated challenges:

Co-elution. Not every LC peak is a single compound. Each extract will be run on several columns with different chemistries, producing different elution profiles. Where compounds still overlap, finely-sampled inline spectra plus curve resolution (MCR-ALS) can often deconvolute them, with the mass spectrometer confirming purity. We note MCR-ALS degrades where congeners share both chromophore and elution behavior; such peaks are flagged, not forced.

Spectral fidelity. We calibrate wavelength against holmium oxide and photometric accuracy against a dichromate standard, rechecking each run. A spectral peak-purity threshold and an upper-absorbance cutoff reject over-concentrated, distorted readings, so only high-confidence measurements enter the atlas.

Solvent environment. A molecule's spectrum depends on its solvent, and under gradient elution every compound elutes into a different mobile-phase composition. Storing metadata (pH, solvent, temperature at elution) makes records auditable but not comparable. We will therefore measure the shift empirically, a standards panel run across the gradient range, so elution conditions can be modeled as a covariate rather than treated as noise.

Detection mismatch. DAD detects at low-ng on-column while MS reaches pg–fg, so we do not expect a spectrum for every MS feature. The atlas deliberately targets the abundant, strongly absorbing subset that carries the chromophores hyperspectral biology cares about.

Annotation rate. Confident structural annotation in untargeted natural-product metabolomics is typically well under 10% of features. We report structure-linked spectra, not total spectra, as our headline metric.

Spectral range. 250–800 nm captures most biologically relevant absorbance and is the off-the-shelf DAD range. NIR-absorbing metabolites are rare but are the highest-value targets for field- and remote-readable reporters; once the methodology is established, an extended 250-1000 nm detector is a straightforward upgrade. The atlas's near-term value for NIR is predictive, learning the structure–spectrum map well enough to design into that window, rather than discovery.