Artificial Intelligence-assisted Biomedical Literature Knowledge Synthesis to Support Decision-making in Precision Oncology

Area of Science:

• Biomedical Informatics
• Computational Biology
• Oncology

Background:

• Effective targeted therapies rely on analyzing tumor molecular profiles and matching them with clinical phenotypes.
• Accessing and synthesizing knowledge from vast biomedical literature, registries, and databases is essential.

Purpose of the Study:

• To evaluate natural language processing (NLP) approaches for knowledge retrieval and synthesis in the biomedical domain.
• To assess the performance of NLP tools in supporting named entity recognition (NER) and relation extraction (RE) for precision oncology.

Main Methods:

• Tested PubTator 3.0, Bidirectional Encoder Representations from Transformers (BERT), and Large Language Models (LLMs).
• Evaluated performance on named entity recognition (NER) and relation extraction (RE) tasks using biomedical texts.

Main Results:

• PubTator 3.0 achieved the highest F1-score (0.93) for NER.
• BioBERT demonstrated superior performance in RE (F1-score 0.79), recognizing most entities and relations.
• BioBERT excelled in a specific use case, identifying nearly all entity mentions and most relations.

Conclusions:

• AI-assisted NLP approaches, particularly BioBERT, significantly enhance knowledge retrieval and synthesis from biomedical literature.
• These findings support the integration of AI tools to facilitate informed decision-making in precision oncology.
• NLP models show promise in accelerating the matching of molecular tumor profiles with clinical phenotypes for targeted therapy selection.