Welcome!

You have found the cui2vec explorer. This website will let you interact with embeddings for over 108,000 medical concepts. These embeddings were created using insurance claims for 60 million americans, 1.7 million full-text PubMed articles, and clinical notes from 20 million patients at Stanford. More information about the methods used to create these embeddings can be found in our preprint: https://arxiv.org/abs/1804.01486

Concept Map
This tab displays an interactive t-sne visualization of all 108,477 concepts.

Concept Similarity
This tab will allow you to sort all concepts by similarity to a query concepts. You can filter by semantic type, which will allow you to answer questions such as "What drug is most related to pancreatic cancer?" or "What symptoms are associated with malaria?".

Availability of pre-trained embeddings

If you would like to download the embeddings, they can be found here .

References

If you use these results in your own work, please cite the following paper:
A. L. Beam, B. Kompa, I. Fried, N. P. Palmer, X. Shi, T. Cai, and I. S. Kohane. 2018. Clinical Concept Embeddings Learned from Massive Sources of Medical Data. ArXiv e-prints (April 2018).

Note: This information provided on this website is for research/educational purposes only and does not constitute medical advice.

Overview

Below is a concept map for each embedding created by cui2vec. The map was created by using t-sne to map the 500-dimensional vectors into 2-dimensional space. The points are colored by the data source in which they were found, and 16,299 points in blue were found in multple sources. The map is fully interactive and will give information on hover and allows you to zoom into certain regions of the plot with your mouse. Clicking on the legend will allow you to hide data from the different sources.

Note: The plot takes a few seconds to load initially, so please be patient.

Overview

Here you can rank all concepts by similarity to a query concept. To search, begin typing your query in the field below and you will be given several suggestions of possible matches. Once you find your desired concept, press the "Go!" button and the results will be displayed in a table to the right. The table is searchable (upper right) and you can filter results using the input fields at the bottom of each column. For instance, if you would only like to see associated concepts that are "Signs of Symptoms", enter this into the field at the bottom of the "SemanticType" column.