CLARINET: efficient learning of dynamic network models from literature.

MOTIVATION

Creating or extending computational models of complex systems, such as intra- and intercellular biological networks, is a time and labor-intensive task, often limited by the knowledge and experience of modelers. Automating this process would enable rapid, consistent, comprehensive and robust analysis and understanding of complex systems.

RESULTS

In this work, we present CLARINET (fying works), a novel methodology and a tool for automatically expanding models using the information extracted from the literature by machine reading. CLARINET creates collaboration graphs from the extracted events and uses several novel metrics for evaluating these events individually, in pairs, and in groups. These metrics are based on the frequency of occurrence and co-occurrence of events in literature, and their connectivity to the baseline model. We tested how well CLARINET can reproduce manually built and curated models, when provided with varying amount of information in the baseline model and in the machine reading output. Our results show that CLARINET can recover all relevant interactions that are present in the reading output and it automatically reconstructs manually built models with average recall of 80% and average precision of 70%. CLARINET is highly scalable, its average runtime is at the order of ten seconds when processing several thousand interactions, outperforming other similar methods.

AVAILABILITY AND IMPLEMENTATION

The data underlying this article are available in Bitbucket at https://bitbucket.org/biodesignlab/clarinet/src/master/.

SUPPLEMENTARY INFORMATION

Supplementary data are available at online.