Brain Reference Architecture (BRA) driven development has been proposed as a means of creating software that learns from and emulates the brain. In BRA-driven development, the Brain Information Flow (BIF), which describes the brain's anatomical structures at a mesoscopic level, is used as a methodology for crafting Brain-inspired software. This process leads to the creation of the Hypothetical Component Diagram (HCD), a virtual function component diagram, serving as the design information for brain-type software.
In order to comprehensively cover the continually growing anatomical knowledge into BIF, we introduce the prompts for automatically extracting anatomical knowledge about the brain from neuroscience literature using large language models(GPT-4), and the Bibliographic Database for BRA (BDBRA) for accumulating the extracted data. The BDBRA is specifically designed as a database focusing on brain region projections, engineered to incorporate functions described by interconnections between brain regions into computational models. In our experiments, we confirmed that the application of appropriate prompts allowed for the extraction of anatomical knowledge of the brain from neuroscience literature(>80% accuracy).Furthermore, to ensure the data recorded in the database accurately captures relevant insights, we examine the credibility of information within neuroscience literature. We also propose a method for evaluating the credibility of such literature.