Recently, the Novellab Lab, under the leadership of Professor Xin Hongyi from the PuYuan Future Technology College at Shanghai Jiao Tong University, has made a remarkable breakthrough. Their paper has been officially accepted for presentation at the renowned International Conference on Machine Learning (ICML) 2026. The paper, titled 'InfoGlobe: Local-and-Global Information-Preserving Statistical Manifold Learning for Single-Cell Transcriptomics', delves into geometric modeling techniques tailored for the analysis of single-cell and spatial transcriptomic data.

Faced with the challenge that current dimensionality reduction methods find it difficult to simultaneously preserve the global geometric relationships between cell states, the Novellab Lab has introduced an innovative statistical manifold learning framework known as InfoGlobe. From an information geometry standpoint, this framework redefines the low-dimensional representations of single-cell data. It effectively maps high-dimensional transcriptomic spaces onto low-dimensional hyperspheres, ensuring the preservation of as much of the original data's information geometry as feasible.

Experimental findings reveal that InfoGlobe surpasses various existing methods in preserving both local and global structures. Moreover, it excels in clustering accuracy, gene program discovery, and the resolution of cell subpopulations.