A Domain-independent Rule-based Framework for Event Extraction

Abstract

We describe the design, development, and API of ODIN (Open Domain INformer), a domainindependent, rule-based event extraction (EE) framework. The proposed EE approach is: simple (most events are captured with simple lexico-syntactic patterns), powerful (the language can capture complex constructs, such as events taking other events as arguments, and regular expressions over syntactic graphs), robust (to recover from syntactic parsing errors, syntactic patterns can be freely mixed with surface, token-based patterns), and fast (the runtime environment processes 110 sentences/second in a real-world domain with a grammar of over 200 rules). We used this framework to develop a grammar for the biochemical domain, which approached human performance. Our EE framework is accompanied by a web-based user interface for the rapid development of event grammars and visualization of matches. The ODIN framework and the domain-specific grammars are available as open-source code.

Methods to Expand Cell Signaling Models Using Automated Reading and Model Checking

“…2a shows examples of these three types of descriptions and the average number of occurrences for each type of interaction in the sample paragraph set. Automated reading engines [2] can extract events in the form of frames that contain an interaction with two entities (arguments). We list in Fig.…”

“…To this end, several automated reading engines have been developed to extract interactions between biological entities from literature. These automated readers are capable of finding hundreds of thousands of such interactions from thousands of papers in a few hours [2]. However, in order to accurately and efficiently incorporate these pieces of knowledge into a model, we need a method to distinguish useful relationships from vast amounts of extracted information.…”

Recipes for Translating Big Data Machine Reading to Executable Cellular Signaling Models

“…The first type of reading engine, REACH [2], can extract both direct and indirect interactions, as well as interaction mechanisms, where available. The simplest, and most common, reading outputs are those that include only a regulated element and a single regulator, each of them having one of the entity types listed in Table 1, and the interaction mechanism being one of the mechanisms described in Table 3.…”

“…We obtain outputs from three types of reading engines, namely REACH [2], RUBICON [26], and Leidos table reading (LTR) [27]. These reading engines provide output files with similar but not exactly the same format.…”

“…When building a model of a particular biological system, one example being cancer microenvironment, researchers usually start by searching for existing relevant models and by looking for information about system components and their interactions in published literature. Although there have been attempts to automate the process of model building [1,2], most often modelers conduct these steps manually, with multiple iterations between (i) information extraction, (ii) model assembly, (iii) model analysis, and (iv) model validation through comparison with most recently published results. To allow for rapidly modeling the complexity of diseases like cancer, and for efficiently using ever-increasing amount of information in published work, we need representation standards and interfaces such that these tasks can be automated.…”