ARMQA: Leveraging GPT and Specialized Models for Accurate Candidate Answer Ranking in Molecule-Based Question-Answering
Keywords: Question Answering, Drug Discovery, LLM, Molecule, QA
TL;DR: This paper presents a novel framework for molecule-based question answering in drug discovery, enhancing answer ranking using predefined candidates, PubMed BERT, and the molecule transformer.
Abstract: Artificial intelligence (AI) has transformative potential in drug discovery. We focus on question answering (QA) for molecules. Existing approaches generate random answers due to the unlimited options. We propose a novel framework using predefined candidate answers. Leveraging Large Language Models (LLMs), we introduce a dataset with one molecule, one question, and multiple candidates. Our approach combines a molecule transformer and PubMed BERT to calculate embeddings for molecules and answers. Each question becomes a separate task. Experimental results show the effectiveness of our method. This research contributes to drug discovery by addressing molecule-based QA challenges and leveraging AI advancements.
Submission Number: 16
Loading