Shumin Li

BLOOM generates developmental cell-expression distributions across time using conditional flow matching, outperforming baselines in gene and cell-expression correlations.

This study shows that estrogen deficiency worsens diastolic dysfunction and energy metabolism in hypertensive female mice, with improvements seen when estrogen or its receptor agonists are administered.

Gungnir is a DNA storage codec that corrects errors through hash matching and biochemical constraints, achieving full data recovery from 20% erroneous bases without redundancy or high-coverage sequencing.

This study evaluates genomic language models' ability to predict personal gene expression using paired genome-transcriptome data, finding limited success in capturing inter-person variability but improved performance for seen genes in new individuals.

AutoPM3 automates PM3 evidence extraction from scientific literature using LLMs, improving variant interpretation for rare diseases.

CellContrast reconstructs spatial relationships in single-cell RNA-seq data using contrastive learning, outperforming existing methods and enabling accurate spatial analysis.

The paper maps the transcriptional regulatory network of Pseudomonas syringae, identifying hierarchical TF organization and functional roles in virulence and metabolism.

ClusterV-Web provides a user-friendly web interface for analyzing Nanopore long-read data to profile HIV quasispecies and generate drug resistance reports.

This study uses ultra-low coverage whole genome sequencing on embryo samples to perform GWAS and identifies genetic loci associated with gestational age.

This paper introduces Clair3, a deep learning-based variant caller that combines pile-up and full-alignment methods, achieving faster performance and better accuracy, especially at lower coverage.