NEW_DATASETS.md

New Datasets Released with RamanBench

This document describes the 17 datasets released for the first time as part of RamanBench, spanning six biological and two chemical dataset groups. All datasets are accessible via the raman-data Python library and published under CC BY 4.0.

Loading Datasets

Install the library:

pip install raman-data

Load any dataset by its raman_data-key:

from raman_data import raman_data

# Load a dataset by raman_data-key
raman_data_key = "kaiser_ecoli_fermentation"
dataset = raman_data(raman_data_key)

# Access data
X = dataset.spectra          # np.ndarray (n_samples × n_wavenumbers)
w = dataset.raman_shifts     # np.ndarray of wavenumber values in cm⁻¹
y = dataset.targets          # np.ndarray of labels or values
target_names = dataset.target_names  # list of target column names

# Convert to pandas DataFrame
df = dataset.to_dataframe()

Overview

	Domain	Task	Datasets	Targets	Samples	Features	Range (cm⁻¹)
E. coli Fermentation (Kaiser)	Biological	Regression	2	8	28	1,699	301–1999
E. coli Fermentation (Time-Gated)	Biological	Regression	2	8	25	114	604–1508
S. thermophilus Fermentation (Kaiser)	Biological	Regression	1	4	14	1,501	300–1800
S. thermophilus Fermentation (Time-Gated)	Biological	Regression	1	4	14	117	487–1423
E. coli Metabolites	Biological	Regression	2	5	2,304	594	402–1599
Bio-Catalysis Monitoring of AXP	Biological	Regression	1	4	344	2,048	−32–3385
Yeast Fermentation	Biological	Regression	1	4	58	1,900	401–2300
R. eutropha Copolymer Fermentations	Biological	Regression	1	6	82	2,776	405–3180
Gasoline Properties (Benchtop)	Chemical	Regression	1	12	179	961	98–3801
Gasoline Properties (Handheld)	Chemical	Regression	1	12	179	1,901	400–2300
Adenine SERS (Colloidal)	Chemical	Regression	2	2	855	534	400–1999
Adenine SERS (Solid)	Chemical	Regression	2	2	2,661	534	400–1999
Total			17	71	6,743

---

E. coli Fermentation (Kaiser)

raman_data-keys: kaiser_ecoli_fermentation, kaiser_ecoli_fermentation_supernatant

Paper: Kogler et al. (2018), doi:10.1002/btpr.2665

Two datasets of E. coli fed-batch fermentation spectra recorded with a Kaiser NIR-Raman spectrometer, capturing both the fermentation broth and the cell-free supernatant. Regression targets are OD600, glucose, acetate, and fermentation time. Samples were collected at defined time points, centrifuged for supernatant preparation, and stored at −80 °C. Reference concentrations were determined by enzymatic assay. Measurements were performed in aluminium microwell plates (20 µL per well).

Instrument: Kaiser RXN1, 785 nm excitation, 135 mW, non-immersion probe (NA = 0.29), 20 s × 5 accumulations, 4 cm⁻¹ resolution, CCD cooled to −40 °C.

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
kaiser_ecoli_fermentation	14	1,699	301–1999	HuggingFace	croissant
kaiser_ecoli_fermentation_supernatant	14	1,699	301–1999	HuggingFace	croissant

---

E. coli Fermentation (Time-Gated)

raman_data-keys: tg_ecoli_fermentation, tg_ecoli_fermentation_supernatant

Paper: Kogler et al. (2018), doi:10.1002/btpr.2665

Two datasets of E. coli fed-batch fermentation spectra recorded with a time-gated Raman spectrometer, capturing both the fermentation broth and the cell-free supernatant. These are the time-gated counterparts to the Kaiser E. coli datasets above; the four targets (OD600, glucose, acetate, fermentation time) and sample preparation protocol are identical.

Instrument: TimeGate TGM1, pulsed 532 nm Nd:YVO₄ laser (~30 mW, 100 ps pulses), fiber-optic probe (NA = 0.22). A temporal gate of 1.2–2.1 ns after the laser pulse suppresses fluorescence from the culture medium. Resolution 10 cm⁻¹, non-cooled SPAD array detector; ~15 min per spectrum.

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
tg_ecoli_fermentation	12	114	604–1508	HuggingFace	croissant
tg_ecoli_fermentation_supernatant	13	114	604–1508	HuggingFace	croissant

---

S. thermophilus Fermentation (Kaiser)

raman_data-key: streptococcus_thermophilus_fermentation_kaiser

Offline Raman spectra from batch cultivations of Streptococcus thermophilus in shake flasks, recorded with a Kaiser RXN1 spectrometer. The dataset covers two independent 24-hour fermentation runs. It is in the tiny-data regime (N = 14) and targets lactose, galactose, lactate, and OD600.

Instrument: Kaiser RXN1, 785 nm excitation (same setup as the E. coli Kaiser datasets above).

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
streptococcus_thermophilus_fermentation_kaiser	14	1,501	300–1800	HuggingFace	croissant

---

S. thermophilus Fermentation (Time-Gated)

raman_data-key: streptococcus_thermophilus_fermentation_timegate

Offline Raman spectra from the same Streptococcus thermophilus batch cultivations as above, recorded with a time-gated Raman spectrometer. Temporal gating suppresses the fluorescence background characteristic of complex fermentation media. Two independent 24-hour fermentation runs are included. Targets are identical to the Kaiser variant (lactose, galactose, lactate, OD600); both datasets are in the tiny-data regime (N < 50).

Instrument: TimeGate spectrometer, pulsed 532 nm laser.

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
streptococcus_thermophilus_fermentation_timegate	14	117	487–1423	HuggingFace	croissant

---

E. coli Metabolites

raman_data-keys: ecoli_metabolites, ecoli_metabolites_dig4bio

Paper: Lange et al. (2025), doi:10.1002/bit.70006

Both datasets contain Raman spectra of aqueous mixtures of key E. coli fermentation metabolites, acquired with an automated high-throughput Raman system integrated into a liquid handling station. ecoli_metabolites covers binary glucose–acetate mixtures; ecoli_metabolites_dig4bio extends the composition to include magnesium sulfate.

Instrument: Metrohm Raman Plus 785 (785 nm, 455 mW), fiber-optic BAC102 probe, 10 s acquisition, 18 µL flow-through quartz cuvette. Samples were robotically pipetted by a Tecan EVO 200 liquid handling station via a multiplexer valve. Concentrations pipetted by the robot served as ground truth, cross-validated against enzymatic assays.

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
ecoli_metabolites	1,920	594	402–1599	HuggingFace	croissant
ecoli_metabolites_dig4bio	384	1,869	402–1599	HuggingFace	croissant

---

Bio-Catalysis Monitoring of AXP

raman_data-key: ht_raman_bio_catalysis_axp

High-throughput Raman spectra for real-time monitoring of biocatalytic reactions involving adenosine phosphates (AMP, ADP, ATP; collectively AXP). The reaction medium uses Deep Eutectic Solvents (DES) as an alternative solvent system and a Tris(hydroxymethyl)aminomethane buffer to fix pH between 7 and 9. When trained as a regression model, the dataset can serve as an analytical method to evaluate the suitability of different enzymes. The four regression targets cover the key phosphorylated forms of adenosine (adenosine, AMP, ADP, ATP).

Instrument: Metrohm Raman Plus 785 (785 nm, 455 mW), fiber-optic BAC102 probe, 25 s acquisition per spectrum, 18 µL flow-through quartz cuvette (same automated measurement platform as the E. coli Metabolites datasets).

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
ht_raman_bio_catalysis_axp	344	2,048	−32–3385	HuggingFace	croissant

---

Yeast Fermentation

raman_data-key: yeast_fermentation

Paper: Legner et al. (2019), doi:10.1002/bit.27112

Online Raman spectra of the continuous ethanolic fermentation of sucrose by Saccharomyces cerevisiae immobilized in calcium alginate beads. Immobilization enables stable continuous processing and unobstructed optical access to the liquid phase. The dataset was originally published without providing access to the raw data. Four regression targets (glucose, fructose, glycerol, ethanol) capture the dynamic evolution of key metabolites during continuous operation.

Instrument: Handheld IDRaman mini 2.0 (Ocean Optics), 785 nm excitation, 400–2300 cm⁻¹, 13 cm⁻¹ resolution, QS 0.5 mm quartz flow cell. Data acquisition automated via MATLAB with cloud upload. Reference concentrations determined by HPLC.

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
yeast_fermentation	58	1,900	401–2300	HuggingFace	croissant

---

R. eutropha Copolymer Fermentations

raman_data-key: ralstonia_fermentations

Paper: Lange et al. (2024), doi:10.1016/B978-0-443-28824-1.50510-X

In-line Raman spectra from four independent batch cultivations of Ralstonia eutropha for biosynthesis of the biodegradable copolymer P(HB-co-HHx). Two cultivations used canola oil as carbon substrate, two used fructose; starting concentrations of RCDW, fructose, and urea were varied to maximise diversity in biomass and metabolite profiles. The dataset combines experimental and high-fidelity synthetic spectra to address multicollinearity between correlated process variables. Six regression targets cover cell dry weight, substrate consumption, and copolymer fractions.

Instrument: Multi-Spec Raman spectrometer (Tec5), 785 nm excitation, up to 500 mW, 365–3180 cm⁻¹, in-line sapphire optical window probe.

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
ralstonia_fermentations	82	2,776	405–3180	HuggingFace	croissant

---

Gasoline Properties (Benchtop)

raman_data-key: fuel_benchtop

Paper: Voigt et al. (2019), doi:10.1016/j.fuel.2018.09.006

FT-Raman spectra of 179 commercial gasoline samples recorded with a benchtop 1064 nm spectrometer. The sample set spans RON 95–102.2 (Super, Super Plus, and premium grades) from refineries and petrol stations. Twelve regression targets include Research Octane Number (RON), Motor Octane Number (MON), ethanol content, oxygenate additives (MTBE, ETBE), and benzene content. Ground-truth RON and MON were measured using a CFR motor per ASTM D2699/D2885. The 1064 nm excitation suppresses fluorescence from aromatic gasoline components that interferes at shorter wavelengths.

Instrument: NXR FT-Raman module (Thermo Fisher Scientific) coupled to a Nicolet 6700 FT-IR spectrometer, 1064 nm excitation, 900 mW, 64 averaged spectra, 8 cm⁻¹ resolution, 2 mL glass vials.

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
fuel_benchtop	179	961	98–3801	HuggingFace	croissant

---

Gasoline Properties (Handheld)

raman_data-key: fuel_handheld

Paper: Legner et al. (2019), doi:10.1021/acs.energyfuels.9b02944

Raman spectra of the same 179 commercial gasoline samples as the Benchtop dataset, recorded with a portable handheld spectrometer. The twelve regression targets are identical, enabling direct evaluation of cross-instrument transferability between laboratory and field form factors.

Instrument: IDRaman mini 2.0 (Ocean Optics), 785 nm excitation, 100 mW, 400–2300 cm⁻¹, 13 cm⁻¹ resolution, 2 mL glass vials.

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
fuel_handheld	179	1,901	400–2300	HuggingFace	croissant

---

Adenine SERS (Colloidal)

raman_data-keys: adenine_colloidal_gold, adenine_colloidal_silver

Paper: Fornasaro et al. (2020), doi:10.1021/acs.analchem.9b05658

Two datasets from a large-scale European interlaboratory SERS study (COST Action Raman4Clinics), using colloidal nanoparticle substrates. Up to 18 European laboratories participated, each using its own 785 nm Raman spectrometer. Citrate-reduced gold and silver nanoparticle suspensions were provided centrally. Aqueous adenine solutions were prepared in PBS (pH 7.4) at multiple concentration levels; the regression target is adenine concentration. The two sub-datasets differ only in substrate metal (Au vs. Ag), enabling comparison of inter-laboratory reproducibility across substrate types.

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
adenine_colloidal_gold	225	534	400–1999	Zenodo 3572359	croissant
adenine_colloidal_silver	630	534	400–1999	Zenodo 3572359	croissant

---

Adenine SERS (Solid)

raman_data-keys: adenine_solid_gold, adenine_solid_silver

Paper: Fornasaro et al. (2020), doi:10.1021/acs.analchem.9b05658

Two datasets from the same European interlaboratory SERS study as above, using sputtered solid substrates instead of colloidal suspensions. Metal-coated nanostructured surfaces (sputtered gold and silver) were distributed centrally to all sites. All other experimental conditions (adenine solutions, PBS preparation, 785 nm excitation) are identical to the colloidal variants. With 2,661 spectra total, this is the larger of the two Adenine sub-collections.

raman_data-key	Samples	Features	Range (cm⁻¹)	Source	Croissant
adenine_solid_gold	810	534	400–1999	Zenodo 3572359	croissant
adenine_solid_silver	1,851	534	400–1999	Zenodo 3572359	croissant