Chemharmony merges many chemical databases in a single set of a 3 tables:
erDiagram
SUBSTANCES {
UUID sid
JSON data
}
PROPERTIES {
UUID pid
JSON data
}
ACTIVITIES {
UUID aid
UUID sid
UUID pid
String source
String inchi
String value
Double numvalue
}
ACTIVITIES ||--o{ SUBSTANCES : activity_has_substance
ACTIVITIES ||--o{ PROPERTIES : activity_measures_property
The primary tables in this project are substances, properties and activites.
Currently, each activity assigns a 0 or a 1 to a given substance and property:
activity ~ substance + property
The 0 or 1 is chosen to indicate whether a given chemical is 'positive' or 'negative' for a given property. The meaning of positive or negative can vary by property. In case of chemical classifications like acute oral toxicity, positive (or 1) means that the chemical is an acute oral hazard and negative (or 0) means that it is not. For numeric properties like LD50 or binding affinity we often choose for positive to mean a chemical hs a value below the median and negative to mean it is above the median. Future versions of this repository will store the raw value of the activity.
There are a few other more experimental tables right now.
Property_categories are the result of a GPT4 classification of the Properties data field, each classification assigns a property to one or more category with an associated reason and strength. A strength of 1 is the weakest and 10 the strongest.
This schema is focused on building QSAR models, but a more flexible graph based schema might be better. Ultimately, capturing data that relates compounds with each other or with other entities through reactions, metabolism, etc. could be captured by adding more tables. Graph schemas can be isomorphic with that approach.