Yang Cui

A deep learning framework based on Temporal Fusion Transformers for jointly forecasting high-frequency vital signs and sparsely sampled laboratory results in ICU settings, implementing novel strategies including sliding window equalized sampling, frequency-aware embedding shrinkage, hierarchical variable selection, and attention calibration across multi-center ICU datasets.

A spatial transcriptomic clustering method leveraging graph convolutional variational autoencoders with self-supervised learning to extract informative latent representations, combined with dual loss functions for robust clustering across diverse tissue datasets.

A framework combining graph convolutional networks with diffusion-based generative models to enhance spatial transcriptomics data quality, enabling improved spatial domain identification and gene imputation even in noise-laden datasets.

A comprehensive clustering toolkit integrating deep learning and machine learning algorithms for versatile analysis of both single-cell RNA-seq and spatial transcriptomics data, providing a unified framework for diverse clustering tasks.

A deep learning framework that integrates gene expression, spatial coordinates, and histological images through graph contrastive learning, enabling accurate spatial domain identification and alignment-free multi-sample integration for spatial transcriptomics data.

A large-scale comparative analysis of approximately 240,000 retinal cells across vertebrate species, identifying 690 regulons governed by 530 regulators and revealing conserved differentiation drivers underlying retinal development and evolution.

A computational study revealing an MHC-II-expressing tumor subpopulation in triple-negative breast cancer with active interactions with immune cells, and establishing a prognostic signature comprising 40 significant genes for clinical outcome prediction.

An automated cell type annotation method utilizing a parallel graph neural network architecture to leverage genome-wide accessibility peak features for accurate and scalable scATAC-seq cell type identification.

An unsupervised adaptive denoising method utilizing a dual-channel joint learning strategy with graph convolutional networks, designed to identify spatial domains and functionally variable genes from noisy spatial transcriptomics data.