Fig. 1: Comprehensive pipeline for tailored sdAb design.

a Traditional sdAb engineering methods involve complex processes, such as combinatorial library construction, screening, and identification. TFDesign-sdAb generates a batch of sdAbs through an integrated generative model, followed by ranking and screening via a scoring model to obtain candidate antibodies. b The generative model of TFDesign-sdAb is based on IgGM, a hybrid-modality diffusion generative model that iteratively generates accurate sdAb sequences and structures through multiple refinement steps. c The ranking model of TFDesign-sdAb is based on A2binder, which integrates pre-trained antigen/antibody large language models and employs a convolutional neural network to predict binding affinity, thereby enabling the ranking of candidate sdAbs.