Biomedical Natural Language Processing using GPU Accelerated Machine Learning

Abstract

Biomedical named entity recognition (BioNER) is a crucial task in biomedical natural language processing (BioNLP) that involves the identification and classification of specific entities in text, such as genes, proteins, diseases, drugs, and anatomical terms. BioNER plays a vital role in various biomedical applications, including information retrieval, literature mining, clinical decision support systems, pharmacovigilance, and the construction of biomedical knowledge bases. However, BioNER faces several challenges, including the handling of domain-specific entities, managing ambiguity and variation, processing large and diverse data, limited labeled training data, resolving ambiguous boundaries, understanding context and semantics, and ensuring scalability and efficiency. Additionally, BioNER algorithms need to be scalable to process large-scale biomedical text data efficiently. This necessitates the development of computationally efficient algorithms and the utilization of parallel computing techniques such as GPU acceleration. Addressing these challenges is crucial for accurately identifying and classifying biomedical entities, enabling effective information extraction and knowledge discovery from biomedical text for research and healthcare applications. newlineIn the first part of this research, an extensive literature review is conducted to identify research gaps in biomolecular event extraction using natural language processing. Machine Learning and Deep Learning methods are then employed, starting with Biomedical named entity recognition and visualization using spaCy. Additionally, a novel Biomedical Named Entity Recognition Tagger called NeRBERT is implemented, which is based on the Bidirectional encoder representation from transformers (BERT) pre-trained language model adapted for biomedical corpora. The research also presents a hybrid approach called Reinforcement Learning Based Distantly Supervised Biomedical Named Entity Recognition, which combines a partial-CRF and a performance-driven, policy-based RL ...