Xinghua Mindy Shi

The paper shows that structural variants from donor-specific assemblies influence chromatin contact divergence in humans, with many SVs linked to regulatory contexts and expression variation.

The paper presents a population-scale Y chromosome reference panel with 142 assemblies, revealing structural and epigenetic variation across 180,000 years of evolution.

This study develops a sample-specific transformer model to improve genotype imputation from sparse single-cell RNA sequencing data, enhancing the recovery of genetic cell types.

The paper assembles and characterizes 2,110 human centromeres, revealing novel haplotypes, mobile element insertions, and mutation rate variation across chromosomes.

Hybrid HiFi/ultra-long ONT assemblies of 65 diverse humans yield 130 haplotypes, 1,246 complete centromeres, and a pangenome panel enabling short-read genome inference at median QV 45.

This paper introduces a Multiple Domain Adversarial Neural Network (MDANN) to improve fairness in disease prediction by reducing biases in brain imaging data for Alzheimer's and Autism populations.

TRUHiC enhances low-resolution Hi-C data using deep generative modeling to improve 3D chromatin structure characterization across cell lines and species.

The paper develops a Hi-C analysis pipeline to catalog TADs and loops in human genomes and quantifies the impact of SVs on chromatin structure and gene expression.

The paper introduces NAHRwhals, a method to detect serial structural variations in long-read genomes, identifies 37 sSV loci in humans, and shows their relevance to disease.

The paper presents long-read assemblies of 43 diverse human Y chromosomes, revealing extensive structural complexity and variation across 180,000 years of evolution.

E2EGraph predicts prostate cancer pathological stages using gene expression data and graph neural networks, achieving state-of-the-art performance with interpretable marker gene identification.

STI is a transformer-based model for genome-scale genotype imputation that achieves high accuracy and handles multi-allelic variants effectively.

FedDP improves disease prediction from imbalanced genetic data by combining federated learning with differential privacy and a weighted minmax loss.

This paper introduces ResActNet, a homomorphic encryption-friendly deep learning network for secure genomics analysis, achieving low approximation errors in classification and regression tasks.

novoRNABreak detects novel splice junctions and fusion transcripts in cancer RNA-seq data with high sensitivity and efficiency.

Mako detects complex structural variants using a graph-based pattern growth approach, achieving 70% validation rate on real data and identifying 15 CSV types including two novel ones.

This study measures RNA secondary structures in Plasmodium falciparum during asexual development, finding that ~71% of structured RNAs are protein-coding and that structural content correlates with translation efficiency.

The paper presents 64 haplotype-resolved human genomes using long-read sequencing and identifies structural variants, SV-eQTLs, and adaptive selection candidates.

This study evaluates differential privacy's effectiveness in defending against membership inference attacks on genomic machine learning models, showing a trade-off between privacy and accuracy.

This study analyzes APE2 genomic alterations and mRNA expression across multiple cancers using TCGA data, finding frequent mutations and upregulated expression in several cancer types.

This study identifies a large number of structural variants in human genomes using multiple sequencing technologies and presents them as a gold standard for future studies.

The paper predicts the effect of NAGLU missense mutations on enzymatic activity using machine learning and PolyPhen-2 scores, finding they are good predictors of functionality.