README.md

Example data (synthetic)

Small synthetic data set for testing the INQUIRE_CSI pipeline without downloading the full source databases. Results obtained with these data have no scientific meaning.

The folder has the same structure as the working folder data/, so its content can be copied into data/ as a whole (see "How to use").

Files

Sample data and household priors

File	Content
Sample_Data.xlsx	500 compounds × 40 synthetic samples plus one blank column (Blank_NIST, listed in dust_cols_drop of Part 1). The compounds are real compounds detected in household dust in the INQUIRE project (InChIKey, name, class). About half of them (253) have at least one source in the example source databases below, and half have none, so both attributed and unattributed compounds occur, as in real samples. Presence/absence was generated at random from each compound's detection frequency, with random sample and sample-group effects. The concentrations are random numbers > 0 (column Concentration Type = "Randomly generated concentrations").
geo_metadata_outputs/household_metadata_priors.csv	Random household likelihoods (0.02–0.98, mid-rank) for the 29 modules, for 30 of the 32 synthetic sample keys (household_id). Two keys are left out on purpose, to test samples without priors.

Example source databases

Each database contains 60 sources (products, foods or emission profiles) in exactly the file format, and with exactly the columns, that Part 1 reads (docs/input_data.md, section 2). CPDat, the literature products and SPECIATE contain one extra test source each (see "Built-in test cases"). In total there are more than 300 sources, because the validation cells of Part 2 draw synthetic mixtures of up to 300 sources.

File	Origin in the pipeline	Sources	Content
Own_Household_Products.csv, Own_Household_Products_Metadata.csv	HOUSEHOLD_PRODUCTS	60	58 synthetic products, up to 4 per sample key for the keys LL01 and LL02 of the sample groups (e.g., AT01_CP01–AT01_CP04), plus 2 real INQUIRE products (FR01_CP04, FR01_CP05). The synthetic products contain compounds detected in the samples with the same key, so they test the preference for a sample's own products.
Literature_Products.csv	LIT_CONSUMER_PRODUCTS	60 (+1 test)	59 synthetic products in 11 consumer-product categories (cleaning products, air fresheners, personal care, toys, flooring, clothing, rubber mats, mosquito repellent), plus 1 real product (CL24, air-conditioner cleaner)
CPDat/product_compositions.tsv, CPDat/products.tsv, CPDat/product_use_categories.tsv	CPDAT	60 (+1 test)	58 synthetic products in 15 product use categories (PUCs), plus 2 real CPDat products (1842, 1843, "Wood Cleaner")
FooDB.csv (tab-separated)	FOODB	60	58 synthetic foods in 16 food groups, plus 2 real FooDB foods (287_01 Alfalfa, 725_44 Snack bar)
SPECIEUROPE.xlsx	SPECIEUROPE	60	58 synthetic profiles in 10 categories (wood burning, traffic, road and urban dust, residential heating, cooking, biogenic, consumer products), plus 2 real profiles (1 Cement, 302 Biomass burning)
SPECIATE.xlsx	SPECIATE	60 (+1 test)	57 synthetic profiles in 12 categories (wood burning, cigarette smoke, cooking, vehicle exhaust, coatings, consumer products, road and soil dust, natural gas, aircraft, refinery), plus 3 real profiles (4663, 4664, 8811)
lookups/source_class_mapping.xlsx	–	–	Example mapping of all categorizations above to the reporting levels source_class ("Domain | Type") and source_class_domain (6 domains). The mapping for the real MCSDB is published on Zenodo together with the databases.

Synthetic sources were built from the compounds of Sample_Data.xlsx: each source contains a random set of compounds (typically 3–20) from compound classes that fit the database (e.g., personal-care compounds and plasticizers in consumer products; fatty acids and flavors in foods; PAHs and petroleum hydrocarbons in emission profiles). Source names and categories use the vocabulary of the real databases, so the priors of Part 2 work as with the real data, but the compositions are random. Concentrations and weight fractions are random numbers; CAS numbers and DTXSIDs are only given for the real example rows.

Real example sources. The few real sources were copied from the processed databases (CPDat v4.0, FooDB v1.0, SPECIATE v5.4, SPECIEUROPE v3.0, INQUIRE household products, and the literature data of Park et al. 2024). They show what real entries look like. Most of their compounds do not occur in the example samples, so some of them are removed by the masking step of Part 1. The terms of use of the original databases apply to these rows (see the main README).

Built-in test cases

Test case	Where	Expected result in Part 1
InChIKey with a digit "0" instead of the letter "O"	Literature_Products.csv, test product CL99	Corrected in the InChIKey cleanup cell (log: "'0' → 'O' corrected: 1 keys")
Junk InChIKeys ("0", "-")	Literature_Products.csv, test product CL99	Set to missing and dropped (log: "junk → missing: 1 rows"; "-" is a placeholder and is set to missing without being counted as junk)
Two products with identical compounds and PUC	CPDat 900001 and test product 900059 (and the real pair 1842/1843)	Merged in the first aggregation (log: "CPDAT: 61 → 59")
The same compound twice in one food	Every synthetic FooDB food (second entry ending in ", free", lower content)	Only the entry with the highest content is kept (log: "Deleted 58 semi-duplicates")
Non-specific aggregate profile	SPECIATE test profile 99999 "Overall Average"	Removed when SPECIATE is loaded
Blank column	Blank_NIST in Sample_Data.xlsx	Removed with DROP_QC_COLUMNS = True
Samples without household priors	2 of 32 sample keys	Listed in the log of Part 2; analyzed with geo multiplier 1.0

Sample IDs

IDs follow the pattern LLNN[_suffix] used by the pipeline (LL = 2-letter sample code, NN = 2-digit sample number): 8 fictitious sample groups (AT, DE, ES, FR, IE, LT, RO, UA) × 4 sample keys (01–04); key 01 of every group has two samples (_LS1, _LS2), e.g., AT01_LS1, AT01_LS2, AT02. The codes are artificial and not linked to real households or to the countries of the INQUIRE study.

What the example data test

• all three notebooks end to end, without downloading the full databases;
• all six source databases, the source class mapping and the reporting levels in Part 3;
• the InChIKey cleanup, the aggregation of identical sources, the FooDB duplicate removal and the masking in Part 1;
• removal of blank columns (DROP_QC_COLUMNS = True);
• household priors (USE_GEO_PRIORS = True), including samples without priors;
• the preference for a sample's own analyzed products (household factor), including several samples per key;
• figures with several sample groups.

How to use

1. Copy the content of example_data/ (all files and the subfolders CPDat/, lookups/ and geo_metadata_outputs/; the README files are not needed) into the working folder data/. If data/ already contains the real databases, use a second working folder instead and change DATA_DIR in all three notebooks.
2. Run INQUIRE_CSI_Pt1_Preprocessing, INQUIRE_CSI_Pt2_Analysis_and_Validation and (optionally) INQUIRE_CSI_Pt3_Visualization as described in the main README, with DROP_QC_COLUMNS = True in Part 1.
3. Part 1 takes a few seconds, Part 2 a few minutes (most of it alpha tuning and validation), Part 3 about a minute.

NB! The example source databases are far smaller than the real ones (~360 instead of ~258,000 sources). Alpha tuning, prior calibration and the validation metrics therefore give values that differ from the reference run (docs/method_and_limitations.md); they only show that the pipeline runs. Expected messages with the example data:

• Part 1: Dust dataset: DF 'dust_strip' not found → skipped (harmless; the sample data are cleaned as "Full dust dataset");
• Part 2, Check the geo mapping: many labels of MODULE_SOURCE_MAP "do not fire" (status NOT FOUND), because the example databases contain only a few source categories;
• Part 2, alpha tuning: the best alpha lies at the lower end of the grid (1000), because the small synthetic sources hardly overlap.

The example data were test-run through Parts 1 and 2 and the heatmap cells of Part 3 when they were created (2026-10-17).

License: CC BY-NC 4.0 (the real example rows keep the terms of their original databases).