Title: Bridging Text and Molecule: A Survey on Multimodal Frameworks for Molecule

Abstract: Artificial intelligence has demonstrated immense potential in scientific research. Within molecular science, it is revolutionizing the traditional computer-aided paradigm, ushering in a new era of deep learning. With recent progress in multimodal learning and natural language processing, an emerging trend has targeted at building multimodal frameworks to jointly model molecules with textual domain knowledge. In this paper, we present the first systematic survey on multimodal frameworks for molecules research. Specifically,we begin with the development of molecular deep learning and point out the necessity to involve textual modality. Next, we focus on recent advances in text-molecule alignment methods, categorizing current models into two groups based on their architectures and listing relevant pre-training tasks. Furthermore, we delves into the utilization of large language models and prompting techniques for molecular tasks and present significant applications in drug discovery. Finally, we discuss the limitations in this field and highlight several promising directions for future research.