Gonzalez Hernandez, Ferran; (2022) Structuring the Unstructured: Unlocking pharmacokinetic data from journals with Natural Language Processing. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Thesis-Ferran-Gonzalez-Hernandez.pdf - Accepted Version Download (5MB) | Preview

Abstract

The development of a new drug is an increasingly expensive and inefficient process. Many drug candidates are discarded due to pharmacokinetic (PK) complications detected at clinical phases. It is critical to accurately estimate the PK parameters of new drugs before being tested in humans since they will determine their efficacy and safety outcomes. Preclinical predictions of PK parameters are largely based on prior knowledge from other compounds, but much of this potentially valuable data is currently locked in the format of scientific papers. With an ever-increasing amount of scientific literature, automated systems are essential to exploit this resource efficiently. Developing text mining systems that can structure PK literature is critical to improving the drug development pipeline. This thesis studied the development and application of text mining resources to accelerate the curation of PK databases. Specifically, the development of novel corpora and suitable natural language processing architectures in the PK domain were addressed. The work presented focused on machine learning approaches that can model the high diversity of PK studies, parameter mentions, numerical measurements, units, and contextual information reported across the literature. Additionally, architectures and training approaches that could efficiently deal with the scarcity of annotated examples were explored. The chapters of this thesis tackle the development of suitable models and corpora to (1) retrieve PK documents, (2) recognise PK parameter mentions, (3) link PK entities to a knowledge base and (4) extract relations between parameter mentions, estimated measurements, units and other contextual information. Finally, the last chapter of this thesis studied the feasibility of the whole extraction pipeline to accelerate tasks in drug development research. The results from this thesis exhibited the potential of text mining approaches to automatically generate PK databases that can aid researchers in the field and ultimately accelerate the drug development pipeline. Additionally, the thesis presented contributions to biomedical natural language processing by developing suitable architectures and corpora for multiple tasks, tackling novel entities and relations within the PK domain.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item