Automated Evidence Retrieval from Literature
Why are we adding AI to our workflow? It’s time to automate the work humans can do, but smart machines can too. Are machines better at these tasks? No, they just free up humans to spend their time on tasks only humans can perform. Who doesn’t want to analyze cases faster, respond to more patients (your boss calls it scaling dry lab operations), or research more?
When we develop AI tools, our goal is to minimize time geneticists spend on routine, repeatable tasks by creating machines smart enough to deliver answers rather than data.
That’s just what the new NLP-based text analytics solution we announced at ACMG19 does. It automatically retrieves the single best piece of evidence from the literature in the context of a case. You can, of course, continue searching the literature for the best or better fit, and even then our search engine presents insights in addition to lists.
Niv Mizrahi, our VP R&D explains it best: “Google has moved beyond 10 blue links, with instant answers and knowledge cards. Our new NLP capabilities do the same in genomics.”
Emedgene’s knowledge base contains 5,500 unique genes with an associated disease, over 25,000 gene-disease connections and 40,000 genes. Much of this data is not available in public DBs and can be identified solely by reading the literature.
The new literature text analytics algorithms were trained on a dataset of 20,000 publications, and achieve a high true positive rate while maintaining a low false positive score. Selecting the correct evidence in the context of a case relies on multiple additional proprietary algorithms.
If you’re interested in seeing it in action and learning more let us know.