Exploration of biomedical knowledge for recurrent glioblastoma using natural language processing deep learning models

Area of Science:

• Computational Biology
• Medical Informatics
• Oncology

Background:

• Recurrent glioblastoma (GBM) treatment knowledge exploration is challenging due to vast data.
• Manual searching of clinical trials and publications is labor-intensive.

Purpose of the Study:

• To apply natural language processing (NLP) and deep learning models to analyze medical research and clinical trial data for recurrent glioblastoma.
• To identify potential therapeutic targets and treatments for recurrent glioblastoma.

Main Methods:

• Fine-tuned the SAPBERT model for question/answering (QA) and named entity recognition (NER) tasks.
• Utilized corpora from Web of Science (2000-2020) and clinicaltrials.gov on recurrent glioblastoma.
• Trained the model on SQUAD2 and custom NER datasets for medical corpora.

Main Results:

• Achieved F1 scores of 0.79 for QA and 0.90/0.76 for drug/gene NER.
• Identified RTK inhibitors and LPA-1 antagonists as promising molecular targets.
• Summarized clinical trials, highlighting bevacizumab, temozolomide, lomustine, and nivolumab.

Conclusions:

• NLP deep learning models can effectively extract knowledge on potential targets and treatments for recurrent glioblastoma.
• The identified knowledge can inform and advance the treatment of recurrent glioblastoma.