Humans have had a long and arduous journey against bacteria. Nowadays, bacterial antimicrobial resistance (AMR) is a serious threat to global health. Under this urgent situation, antimicrobial peptides (AMPs) have been highly anticipated by scientists as potential alternatives to antibiotics. However, numerous AMPs suffer from weak antibacterial activity, unknown toxicity status, and easy inactivation during production and transportation, making it difficult to apply them widely.

In March 2025, Prof. Tingjun Hou and Prof. Changyue Xie from CPS-ZJU, together with the team of Director Zhihui Jiang from Southern Medical University, published a paper in Science Advances entitled “Discovery of antimicrobial peptides with notable Discovery of antimicrobial peptides with notable antibacterial potency by an LLM - based foundation model”. The team proposed AMP-Designer, an LLM-based foundation model, to design 18 broad-spectrum and efficient AMPs in only 11 days by integrating cue learning, model distillation, and reinforcement learning with the help of a LLM model, which is expected to play an important role in combating bacterial resistance in the future. The results are expected to play an important role in combating bacterial drug resistance in the future.

Figure. 1. The overview of AMP-Designer.

The CPS-ZJU is the first author of this paper, Dr. Jike Wang, Associate Professor Yu Kang and Jianwen Feng of Southern Medical University are the co-first authors, and Professor Tingjun Hou and Professor Changyue Xie of Zhejiang University and Director Zhihui Jiang of Southern Medical University are the co-corresponding authors.

Original link: https://doi.org/10.1126/sciadv.ads8932