Expanding the Understanding of Population Descriptors for Biomedical Research to Low- and Middle-Income Countries (LMIC) - Corpora Development

Transformation of biomedical research through meticulous development and expansion of resources for curation, extraction and translation for clinical interpretation is a fundamental component of clinical genetic practices. In clinical contexts, curation of molecular and clinical entities from available sources provides real-time data collection, interpretation and reporting. There is an increasing need to expand the current practices to include population descriptor entities. However, populations can be described in various ways including social, biological and geographical constructions. Furthermore, these constructs can differ across both place and time, resulting in multiple dynamic population entity types. Here, we aim to develop an event annotated corpora for the task of extracting population descriptors. In our approach, we include identification of population target types and guidelines for population entity annotation. We extend the approach to biomedical information extraction to include all types of population descriptors. In development of this corpus, we manually annotated biomedical literature using a combination of structured and unstructured representation for entity extraction across multiple trained annotators. Our event extraction approach was applied to a variety of population descriptor extraction targets, including clinical, demographic, geographic and social. With a focus on Low- and Middle-Income Countries (LMICs), we selected 100 biomedical research papers from biomedical journals published in LMIC countries marking references to population descriptor entities and domain-relevant processes. Trained annotators evaluated relevant entity types for annotation and developed a set of detailed guidelines for annotation in text. Secondly, experts created structured event annotation in both abstracts and published manuscripts. The resulting corpora is intended to serve as a reference for FAIR training and evaluation methods for population entity mention detection in biomedical science from LMICs.