Consistent: 57 of 59. Families with strong, well-separated bands (PE, PP, PS, PET, PMMA, nylon, PAN, PC) come out as expected. PVC without a visible C–Cl band and PVA with residual acetate are now recognised. Two spectra are still wrong: EVA-3 (a blend with aromatic bands) and EVOH-2 (real amide I and II bands).
Acceptable. Very low contrast (~5 %T). The engine returns aliphatic alcohol, which is acceptable for cellulose; it does not separate lignin from hemicellulose.
Partial: 3 of 5. Cell wall: broad O–H, polysaccharide C–O at ~1038 and a β band at ~891. Amide II was not detected, so it does not appear as protein. Three spectra (control, metals 1× and 2×) come out as carbohydrate; the highest metal concentrations (4× and 8×) still come out as aliphatic alcohol because the broad C–O band is relatively weak, and two also show “carbonyl” at ~1635, which is the water bend or amide I. Not yet treated.
Correct: sees only water. These are water spectra. The engine is right, but the RNA bands sit below ~0.05 absorbance under the water. An aqueous-sample warning is under study.
Good: 291 of 291 as amides. Protein (haemoglobin) with amide I (~1646), amide II (~1532) and amide III (~1240), on a diamond ATR. The engine also reads the raw OPUS files (301 tested; 295 with amides) without conversion.
Dybas, J. et al. FTIR spectra, secondary structure alterations of RBC assessed by FTIR-ATR in ApoE/LDLR-/- mice. Figshare, 2023. doi:10.6084/m9.figshare.22155080 · CC BY 4.0
HA / PVA hydrogels
Consistent
4 spectra · example: HA
The engine returned
CarboxylateHydroxyl or N–HPossible inorganic material
Consistent: 4 of 4. The Y axis comes as a fraction (0–1) and the engine works out by itself that it is transmittance. It gets the carboxylate and the polysaccharide of hyaluronic acid, and PVA with residual acetate now comes out as ester and alcohol.
Consistent: 10 of 10. The strongest starch band, at ~995 cm⁻¹, sat outside the range the carbohydrate rule looked at and none of the 10 received a class; today all 10 come out as carbohydrate (7 with the specific polysaccharide class and 3 with the generic one). A sample with a shoulder at ~1148 came out as aliphatic alcohol until the rule also began to consider the width of the C–O band, much broader in starch than in short-chain alcohols. Two carry a mixture note because of a real weak carbonyl at ~1734 (lipid or acetyl).
Mardhotillah, A. B. A. et al. FTIR and GC-MS metabolomics of ten Indonesian tubers. Figshare, 2026. doi:10.6084/m9.figshare.33061928 · CC BY 4.0
Pea starch during germination
Consistent
5 spectra · example: Isolated starch
The engine returned
Polyhydroxylated alcohols and carbohydrates
Consistent: 5 of 5. Same ~995 cm⁻¹ band as the tubers: before, 0 of 15 starch spectra had a class; today the 5 pea spectra come out as polysaccharide. Germination barely changes the spectrum.
Consistent. Recognises the inorganic material and no longer adds alcohol (the Si–O–Si at ~1090 is not C–O) or carbonyl (the water at ~1637 is not C=O).
Partial: 4 of 6. The Si–O (~1086) no longer turns into alcohol in 4 of the 6 columns. In those carrying the OTS alkyl chain (CO and NO, identical in the file received: to be checked against the original) the class still comes out as aliphatic alcohol because of the C–H, and the functionalisation (APTES, OTS) is not described.
Outside the design range. The files cover only ~1151–3499 cm⁻¹. Without the fingerprint region, 116 of 159 get no class, and the engine owns that limitation: it flags the data as inadequate (evidence level 1) and describes the bands without inferring a class.
McDonough, L. FTIR of soils, CaCO3, sucrose, IHSS standards, ash and vegetation. Figshare, 2024. doi:10.6084/m9.figshare.25323205 · CC BY 4.0
Declared limits
Anhydrides: 0 of 5. The engine has no rule for them yet; they come out as acid, acyl chloride or no class.
Haloalkanes with CH₂: 0 of 3, also without a rule.
Ureas have no class of their own: three of the four come out without “amide”.
Liquid alkylphenols: 1 of 4 as phenol.
Alkenes as a liquid film: 3 of 7.
False “carbonyl” in 3 liquid-film alkenes: the C=C band at ~1640 is read as carbonyl, the same pattern seen in the yeast (~1635).
How to read this compilation
Curves are drawn as relative transmittance (bands pointing down), normalised between the lowest and highest value of each spectrum. The original data stay with the authors; here we only show the engine's reading. Note: the starch, silica, hydrogel and polymer cards show data that helped correct the rules; they are a demonstration, not independent validation. Items “under review” still have a correction planned.
Credits and licences
All datasets are open and used with credit to the authors. The identifiers below lead to the original dataset.
Dybas, J. et al. FTIR spectra, secondary structure alterations of RBC assessed by FTIR-ATR in ApoE/LDLR-/- mice. Figshare, 2023. https://doi.org/10.6084/m9.figshare.22155080. Licence CC BY 4.0.