European Commission logo
English English
CORDIS - EU research results
CORDIS

Deep learning-based text mining for interpretation of omics data

Project description

Novel text-mining technology for interpretation of omics data

The omics technologies produce Big Data at an increasingly high rate, and their interpretation involves an association between individual entities in the context of molecular networks. These associations are derived, not only from the omics data but importantly, the pre-generated networks created by text mining of millions of scientific articles. The EU-funded DeepTextNet project aims to extract novel information from biomedical literature sources on the type and direction of molecular associations. Specifically, the objective is to build a next-generation text mining technology for relation extraction of molecular interactions that utilises deep learning and uses Big Data for training, as opposed to small manually curated datasets used in current methodologies.

Objective

"The academic community and the pharmaceutical industry use omics technologies to produce big data at an incredibly increasing rate but are faced with major challenges when it comes to their interpretation. Key for this interpretation is the association between individual entities, which in a biological context means creating molecular networks. These associations cannot be derived from the omics data alone, but rely heavily on pre-generated networks created by text mining of millions of scientific articles. One of the most popular sources of such networks is the STRING database, which currently serves ~100,000 users monthly.
Many of these users work with omics data and a major obstacle, which limits potential benefits for them, is that literature-derived networks are made up of ""functional associations"", stating only that two molecules do something together, but neither the interaction type nor the direction. Hence, our hypothesis is that state-of-the-art computational approaches will be able to exploit new possibilities in network biology that emerge from big data. The key objective of DeepTextNet is to extract novel information from the biomedical literature on the type and direction of gene/protein associations. Specifically, a new paradigm will be realized by building a next generation text mining technology for relation extraction of molecular interactions that explicitly utilizes deep learning and, in contrast to current methodology, makes use of big data for training as opposed to small manually curated datasets. This new strategy for obtaining comprehensive molecular networks with both type and direction for the interactions is precisely what is currently missing for the interpretation of omics data. We expect the impact to be high and wide, as on top of applying this strategy on omics datasets as part of the project, the new technology will feed directly into STRING, which is used globally and integrated into workflows in both academia and industry."

Coordinator

KOBENHAVNS UNIVERSITET
Net EU contribution
€ 207 312,00
Address
NORREGADE 10
1165 Kobenhavn
Denmark

See on map

Region
Danmark Hovedstaden Byen København
Activity type
Higher or Secondary Education Establishments
Links
Total cost
€ 207 312,00