A new AI based platform incorporating machine learning, natural language processing and knowledge graphs, makes biological and biomedical data accessible and consumable. This tool will speed up the discovery of e.g. drugs, green chemicals and other sustainable products.
Often, scientific data is made available in papers, printed in graphs or shared in other unstructured formats. This makes the data impossible for computers to ‘consume’ and integrate with other data sets.
“If this data can be read, stored as structured knowledge, and cross-connected with other data sources, the data would ‘come alive’ and provide real value to the world,” says Chief Information Officer at The Novo Nordisk Foundation Center for Biosustainability Evelyn Travnik.
So, this is what she and her Informatics team at The Novo Nordisk Foundation Center for Biosustainability at Technical University of Denmark (DTU) and University of California, San Diego (UCSD) is now doing: Building an artificial intelligence system called Lifelike that can interpret hidden, inaccessible and ‘dead’ data that is typically stored in Power Point slides, Google Slides, electronic lab books and scientific papers. Lifelike can quickly and in a very user-friendly way understand unstructured text, extract key information in a visual format and cross-connect it with other sources of knowledge – and with the use of machine learning provide deep insights.
"This platform will change the way we share our scientific output with the world - enabling interoperability and reusability"
Evelyn Travnik, Chief Information Officer
“We want Lifelike to be the new virtual assistant that can give scientists the possibility to find hidden biological connections in seconds rather than in months or years. The features in Lifelike will lead to quicker discovery of sustainable solutions – that is our wish and hope,” Evelyn Travnik concludes.
Looking for collaborators
Lifelike is still being developed, but a first production release is now available. This version allows for identification of e.g. organisms, genes, proteins, diseases and custom terms from PDF files and other ‘dead paper’ formats. It enables the import of experimental data and knowledge enrichment across multiple internal and external databases and sources, creating a 360-degree view of the data. Lifelike additionally provides the researcher with the ability for visual notetaking and knowledge reconstruction by dragging entities and text onto a canvas and creating new connections.
At this point, it is very important to get more experts on board with the same vision, Evelyn Travnik explains:
“We really want to get the best people from all over the world onboard – people who share the same vision. The partners could be from academia, but indeed also industrial partners who would like to establish a next generation informatics infrastructure and move towards explainable AI.”
- Annotate any PDF to automatically identify organisms, genes, proteins, diseases, phenotypes etc. with common identifiers and links to online information resources;
- Create new annotations on the fly and associate them with identifiers;
- Drag annotated entities onto a canvas and draw relationships between nodes to capture new knowledge;
- Consolidate large amounts of knowledge into a big-picture view with links to detailed information;
- Import experimental data, simulation results, and NLP-based relationship extractions to view them as an interconnected graph, integrate them with knowledge from public databases to contextualise and enrich the data;
- Create knowledge maps in a collaborative fashion and share with anyone either online or as an exported PDF; and
- Browse and query a global network of biomedical relationships