An uncertainty-graded property matrix
Search 16,000 compounds by property.
16,116 drugs, natural compounds, research chemicals, venoms, and extracts, each rated on 130 structural, chemical, pharmacological, and clinical attributes and summarized here on 11 dimensions. Describe the profile you are looking for, or start from compounds you already know.
- Compounds
- 16,116
- Dimensions
- 11
- Scale
- −4 to +4
About the resource
A property matrix for more than 16,000 compounds
PharmaDive is a dataset covering more than 16,000 substances — drugs, natural compounds, research chemicals, venoms, extracts, and more — each rated on 130 structural, chemical, pharmacological, and clinical attributes. Instead of pulling facts from documents, it has a web-search-enabled language model judge every attribute for every compound on a graded scale from "definitely no" to "definitely yes".
Just five years ago, a project like PharmaDive would have been nearly impossible without a generous, multi-year grant. With today's A.I. capabilities, Tyler Moore was able to build PharmaDive with about $7,000 in Grok (grok.com) API costs. The manuscript describes the data and methods in full, and the pipeline code can be used for your own work.
Suggested citation
Moore, T. M. (2026). PharmaDive: A Language Model Generates an Uncertainty-Graded Property Matrix for More Than 16,000 Compounds. Retrieved from https://pharmadive.org.
Search
Two ways to query the resource
How it works
How the ratings are made
-
1
Every compound is rated
A web-search-enabled language model judged each of the 16,116 compounds on 130 structural, chemical, pharmacological, and clinical attributes, on a graded scale from "definitely no" to "definitely yes". The full matrix is available for download above.
-
2
11 summary dimensions
A psychometric model condenses those ratings into 11 dimensions, each scaled from −4 to +4 around the average compound. The sliders on this page set targets on those same scales.
-
3
Two ways to query
Describe a property profile returns the compounds whose profile lies closest to your settings. Start from example compounds returns those whose full rating profile correlates most strongly with the examples you named.
The data behind PharmaDive is available through the MooreMetrics API.