Biology subjects

Bei, Z.

Publications and source records attributed to Bei, Z..

2 recordsLinked to original sources

bioRxiv2025

MIMYR: Generative modeling of missing tissue in spatial transcriptomics

Spatial transcriptomics enables the study of how gene expression is organized across tissues, revealing how cells interact within their native microenvironments in health and disease. However, tissue damage during sectioning and the allocation of intermediate slices to other assays often result in regions or entire planes missing from the data, limiting downstream analysis. Here, we introduce MO_SCPLOWIMYRC_SCPLOW, a generative framework for reconstructing realistic spatial transcriptomics data in unmeasured tissue regions. MO_SCPLOWIMYRC_SCPLOW addresses this challenge through three coupled components: predicting cell locations via guided diffusion, assigning cell types through supervised classification, and generating gene expression profiles with a transformer conditioned on spatial and cellular context. MO_SCPLOWIMYRC_SCPLOW accurately reconstructs held-out regions in mouse brain data and generalizes across experimental conditions, including variations in gene panels and slicing orientations. After finetuning on limited Alzheimers disease data, MO_SCPLOWIMYRC_SCPLOW captures disease-associated transcriptional changes in unmeasured brain regions. By enabling high-fidelity spatial imputation from limited training data, MO_SCPLOWIMYRC_SCPLOW extends the utility of spatial transcriptomics, allowing researchers to recover unmeasured tissue states and deepen investigations into tissue spatial organization and dynamics.

bioinformatics↗

bioRxiv2024

MSAGPT: Neural Prompting Protein Structure Prediction via MSA Generative Pre-Training

Multiple Sequence Alignment (MSA) plays a pivotal role in unveiling the evolutionary trajectories of protein families. The accuracy of protein structure predictions is often compromised for protein sequences that lack sufficient homologous information to construct high-quality MSA. Although various methods have been proposed to generate virtual MSA under these conditions, they fall short in comprehensively capturing the intricate co-evolutionary patterns within MSA or require guidance from external oracle models. Here we introduce MSAGPT, a novel approach to prompt protein structure predictions via MSA generative pre-training in the low-MSA regime. MSAGPT employs a simple yet effective 2D evolutionary positional encoding scheme to model the complex evolutionary patterns. Endowed by this, its flexible 1D MSA decoding framework facilitates zero-or few-shot learning. More-over, we demonstrate that leveraging the feedback from AlphaFold2 can further enhance the models capacity via Rejective Fine-tuning (RFT) and Reinforcement Learning from AF2 Feedback (RLAF). Extensive experiments confirm the efficacy of MSAGPT in generating faithful virtual MSA to enhance the structure prediction accuracy (up to +8.5% TM-Score on few-shot scenarios). The transfer learning capabilities also highlight its great potential for facilitating other protein tasks.

bioinformatics↗