Automated Formalization of Biological Model Properties into Temporal Logics Using Large Language Models

EasyChair Preprint 11514

Abstract

 In this paper, we demonstrate for the first time that large language models (LLMs) can be used to translate descriptions of biological model properties into formalized linear temporal specifications (LTL). We obtain these properties from published work on biological models and then use GPT-3.5 and 4 to formalize the description using LTL. Previous work decomposes the problem into multiple steps and utilizes multiple translation algorithms to perform the conversion. This decomposition of the translation task was needed with older neural networks and LLM models but is non-intuitive and can lead to compounding the accumulated errors. Our experimental evaluations show that state-of-the-art LLMs such as GPT-3.5 and 4 can successfully generate model specifications from descriptions without decomposing the translation into multiple sub-tasks and can provide an intuitive and convenient way to convert natural language into LTL specifications.

Keyphrases: Computational Systems Biology, LLMs, biological models, formal models, temporal logics

@booklet{EasyChair:11514,
  author    = {Sumit Kumar Jha and Pranav Sinha and Sunny Raj},
  title     = {Automated Formalization of Biological Model Properties into Temporal Logics Using Large Language Models},
  howpublished = {EasyChair Preprint 11514},
  year      = {EasyChair, 2023}}