Antiviral peptides (AVPs) represent a novel and promising therapeutic alternative to conventional antiviral treatments, due to their broad-spectrum activity, high specificity, and low toxicity. The emergence of zoonotic viruses such as Zika, Ebola, and SARS-CoV-2 have accelerated AVP research, driven by advancements in data availability and artificial intelligence (AI). This review focuses on the development of AVP databases, their physicochemical properties, and predictive tools utilizing machine learning for AVP discovery. Machine learning plays a pivotal role in advancing and developing antiviral peptides and peptidomimetics, particularly through the development of specialized databases such as DRAVP, AVPdb, and DBAASP. These resources facilitate AVP characterization but face limitations, including small datasets, incomplete annotations, and inadequate integration with multi-omics data.The antiviral efficacy of AVPs is closely linked to their physicochemical properties, such as hydrophobicity and amphipathic α-helical structures, which enable viral membrane disruption and specific target interactions. Computational prediction tools employing machine learning and deep learning have significantly advanced AVP discovery. However, challenges like overfitting, limited experimental validation, and a lack of mechanistic insights hinder clinical translation.Future advancements should focus on improved validation frameworks, integration of in vivo data, and the development of interpretable models to elucidate AVP mechanisms. Expanding predictive models to address multi-target interactions and incorporating complex biological environments will be crucial for translating AVPs into effective clinical therapies.

抗病毒肽 （AVP） 因其广谱活性、高特异性和低毒性而成为传统抗病毒治疗的一种新颖且有前途的治疗替代方案。在数据可用性和人工智能 （AI） 进步的推动下，寨卡病毒、埃博拉病毒和 SARS-CoV-2 等人畜共患病毒的出现加速了 AVP 研究。本综述重点介绍了 AVP 数据库的开发、其理化性质以及利用机器学习进行 AVP 发现的预测工具。机器学习在推进和开发抗病毒肽和肽模拟物方面发挥着关键作用，特别是通过开发 DRAVP、AVPdb 和 DBAASP 等专业数据库。这些资源有助于 AVP 表征，但面临局限性，包括数据集小、注释不完整以及与多组学数据整合不足。AVP 的抗病毒功效与其物理化学特性密切相关，例如疏水性和两亲性α螺旋结构，这使得病毒膜破坏和特异性靶标相互作用成为可能。采用机器学习和深度学习的计算预测工具显著推进了 AVP 发现。然而，过拟合、有限的实验验证和缺乏机制见解等挑战阻碍了临床转化。未来的进展应侧重于改进的验证框架、体内数据的整合以及开发可解释的模型以阐明 AVP 机制。扩展预测模型以解决多靶点相互作用并整合复杂的生物环境对于将 AVP 转化为有效的临床疗法至关重要。