A zero-shot NLP-based pipeline for automated processing of antimicrobial-related scientific texts

¹ PhytoMicrOmics - Max Planck Tandem Group, Faculty of Engineering, Universidad Nacional de Colombia, Bogotá 111321, Colombia
² MindLab Research Group, Faculty of Engineering, Universidad Nacional de Colombia, Bogotá 111321, Colombia

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Information extraction from literature is a fundamental process in the construction of knowledge in the life sciences. However, it is also a process that often requires time and effort to obtain accurate results. This work proposes a fast and adaptable scheme for the automatic processing of article texts (abstracts) based on the use of NLP models, specifically designed to identify publications related to the evaluation of antimicrobial compounds. The proposed mechanism receives an abstract as input and determines whether the article meets a series of criteria, also generating a list of the chemical compounds present in the text. The NLP models applied to the texts are executed without additional training (zero-shot learning), and as many filtering criteria as necessary can be used. The quality of this proposal is determined by its use in 368 abstracts of articles, employing three acceptance criteria. The results indicate a high precision of the proposed mechanism for both classifying texts in the area of antimicrobial prospecting and recognizing chemical entities.