VDOS2024 Talk#3 Hasin Rehana - Cancer Vaccine Adjuvant Name Recognition from ~ скачать в хорошем качестве

VDOS2024 Talk#3 Hasin Rehana - Cancer Vaccine Adjuvant Name Recognition from ~

VDOS 2024: The 13th International Workshop on Vaccine and Drug Ontology Studies. Workshop Website (Virtual) https://vdos-workshop.github.io/vdos2... Presented by Ms. Hasin Rehana, University of North Dakota. Abstract: "Adjuvants are substances added to vaccines to modify and improve their effects, either by enhancing the body's immune response or by reducing side effects. By stimulating the immune system and increasing antibody production, adjuvants significantly boost the efficacy of cancer vaccines. Identifying adjuvant names in cancer vaccine clinical trial data is essential for advancing research and improving treatment outcomes. However, manually curating adjuvant names from the rapidly growing biomedical literature is challenging. This study investigates the automatic identification of vaccine adjuvant names using Generative Pretrained Transformers (GPT), a large language model (LLM), a form of artificial intelligence. We used state-of-the-art language models, specifically GPT-4, to tackle these challenges. We analyzed two distinct subsets of cancer vaccine trials from https://clinicaltrials.gov/. The first subset included 97 clinical trial records annotated by the researchers of the AdjuvareDB website, serving as the gold standard. GPT-4 achieved an impressive F1-score of 81.9% on this dataset. The second subset consisted of 430 cancer vaccine clinical trials, manually curated by our team to cover a diverse range of adjuvant information and their contextual applications. GPT-4 reached an F1-score of approximately 81.0% on this dataset, with a precision of 92.5% and a recall of 72.1%. From this set, we identified eight new adjuvants not included in the AdjuvareDB, such as Matrix-M1, GLA-SE, and Imiquimod. Our key findings show that GPT-4 is particularly good at recognizing adjuvant names, even rare and novel ones, which are often difficult for traditional named entity recognition approaches. The model's high precision indicates its effectiveness in correctly identifying adjuvant names. However, there is room for improvement in capturing all relevant adjuvant names. Additionally, the model showed robustness in distinguishing adjuvant names from other biomedical entities, reducing false positives and improving overall reliability. In conclusion, this study highlights the potential of Generative Pretrained Transformers in advancing cancer vaccine research through accurate and efficient adjuvant name recognition from clinical trial data. By leveraging the power of GPT-4, we demonstrate a promising approach that bridges the gap between unstructured clinical trial data and actionable insights, ultimately contributing to the progress of cancer immunotherapy. Future studies will focus on expanding beyond cancer vaccine adjuvants to include vaccines for infectious diseases, and comprehensively processing all records from clinicaltrials.gov instead of subsets. Additionally, we aim to extend our search to biomedical literature in PubMed and PubMed Central. Technically, we plan to utilize open-access large language models, such as LLaMA3 and Google's models, to fine-tune our models with sentences related to vaccines and adjuvants, enhancing their generalizability and effectiveness across various contexts."