Neurology in the ChatGPT Era: Artificial Intelligence Advances in Literature-Based Discovery*
Time: 12:00 PM to 1:00 PM
Description
Keeping up with the latest scientific literature in one's own niche specialty can be daunting – much less keeping up with the millions of other scientific articles that may be important to a cross-cutting field like neurology. Recent advances in natural language processing and large language models, including ChatGPT, have greatly increased the accessibility to large volumes of clinical and scientific research. The ability to integrate, aggregate, extract, and summarize data can greatly expedite the speed and efficiency of neurology research. In particular, natural language processing and the use of knowledge graphs has been exciting for the field of drug repurposing, including finding adjuvant therapies for neurological disease. Large language models like ChatGPT open the door to automating the tedious task of clinical meta-analysis and can potentially provide state-of-the-art, statistics-based guidelines in a fraction of the time. In addition to the two talks, there will be an interactive discussion on related relevant easy-to-access literature-based discovery tools for everyday use by neurologists and neuroscientists.
Objectives
Provide overview and interactive demonstration of natural language processing (NLP) and large language models (LLM) for applications in neurology research.
Provide overview and interactive case studies for the use literature based discovery and related text relationship mining tools for neurology research.
Illustrate how NLP and LLM can be used to extract, organize, and mine large-scale, multi-factorial literature from across domains.
Use literature-based discovery to demonstrate specific case studies in glioblastoma, Parkinson's Disease, and other neu diseases will be used to show the utility of mining 34+ million PubMed articles for drug repurposing, adjuvant drug discovery, novel drug targets, and spectral disease mechanism identification.
Provide detailed standard-of-care clinical practice example for automated meta-analysis to determine aggregate effect sizes across multiple cohorts.
Cross-Domain Natural Language Processing to Redefine Disease and Drug Targets: A Case Study with Parkinson's
DescriptionThe scientific literature provides a wealth of information that is siloed by specialty domain and journal. The ability to stitch relationships from over 33 million PubMed articles presents an incredible opportunity for drug repurposing and novel drug targeting. Here we will present recent open-source tools for cross-domain text mining that are relevant to all of neurology. Specifically, advances in artificial intelligence-enabled knowledge graphs are explored. Knowledge graphs stitch together multifactorial and multi-scalar heterogeneous semantic relationships from journal article texts using the Unified Medical Language System (UMLS) ontology. A case study is presented where unsupervised machine learning with the open-source SemNet algorithm was able to identify and rank the most promising repurposed drug candidates for Parkinson’s Disease using text from 33 million journal articles. Primary results showed that antihistamines like ebastine and levocetirizine were promising adjuvant Parkinsonian therapies capable of reducing oxidative stress, improving neurotransmitter balance, and decreasing inflammatory mediators. Moving forward, the ability to integrate knowledge graphs with large language models will be powerful for pre-screening and expediting repurposed and novel therapeutics for neurological disease.
SpeakersLarge Language Models Towards the Automation of Clinical Meta-Analysis: A Case Study with Glioblastoma
DescriptionSystematic literature reviews (SLRs) and meta-analyses form the backbone of gold-standard clinical practice by establishing evidence-based guidelines for patient treatment. Unfortunately, these reviews each consume hundreds to thousands of hours of clinician time to screen studies curate question-specific data buried in the manuscripts of articles. Using a case study on glioblastoma, we demonstrate how large language models can assist clinicians in identifying relevant literature, filtering studies for relevance, and extracting structured clinical data. We use these methods to create a database of treatment protocols and clinical outcomes supporting glioblastoma and other types of cancers. We conclude with practical resources to enable clinicians to use AI to provide evidence-backed insight into open clinical questions.
Speakers