aroma-flavor-chemicals-db

Extract data on aroma & flavor chemicals to a database.

Purpose

This repository houses a database of food chemicals with aroma and/or flavor data. It includes Jupyter notebooks showing the extraction of that data from reference sources, as well as the data itself.

Contents

This repository includes:

• Jupyter notebooks in the root directory:
  • 1-extraction-fenaroli.ipynb: Crude data parsing of aroma/flavor data from a PDF copy of Fenaroli's Handbook of Flavor Ingredients.
  • 2-extraction-common-materials.ipynb: Crude data parsing of additional aroma/flavor data from a PDF copy of Common Fragrance and Flavor Materials.
  • 3-nlp-cleaning.ipynb: A natural language approach to further cleaning the descriptions extracted in the first two notebooks. Broad categories for flavor/smell descriptors are defined, and molecules are assigned to one or more categories.
  • 4-data-validation.ipynb: Validation of chemical names from notebook 3 and lookup of chemical structures. Both steps use the PubChem database.
  • 5-chemical-space-network.ipynb: Application of the data set to create a visualization to answer the question: what types of structural features tend to smell floral?
• Data files in the processed-data directory:
  • aromas-flavors.db: A SQLite database containing the extracted flavor and aroma data. Included tables:
    • fenaroli: Crudely extracted data from Fenaroli's Handbook. Each row represents one molecule, and the included fields are: the chemical name, its unique CAS_num, and a full_description. This data table is used only for storage of partially processed data and is not intended as a primary source of readily usable information.
    • common-materials Similar to above, except with data extracted from Common Fragrance and Flavor Materials. Also as above, this table is not intended as a primary source of readily usable information.
    • categorized-descriptors: Cleaned & well structured sensory information. With one row per molecule, fields are: a molecule's common_name, a list of categories describing its aroma, and a list of categories describing its flavor. (Categories are defined in lexicon.csv, below).
    • structures: Chemical structures for all the molecules present in the categorized-descriptors table. With one row per molecule, fields are: the molecule's common_name and its chemical structure as both a SMILES_code and an InChI_code.
  • Several CSV files:
    • lexicon.csv: A set of 29 categories/subcategories generated to describe flavors and aromas for this project.
    • stems-to-lexicon.csv: A translation dictionary that equates Snowball-stemmed sensory descriptors to categories defined in the lexicon. For example, "rosy" is assigned to the category "floral".
    • aromas-flavors-categorized.csv: Tab-separated file to avoid convolution by molecule names that also contain commas. Each entry describes one molecule by its name, its unique CAS_num identifier, and lists of categories from the lexicon describing both its aroma and flavor.
• floral-chemical-network.png: The cover image used above. Its creation is detailed in the Jupyter notebook 5-chemical-space-network.ipynb.