Yan Cui

Where Tabular Foundation Models Falter on Genetic Data: Datasets That Expose and Provide a Path to Address the Gap

The paper evaluates tabular foundation models on genetic data, identifies ancestry-dependent non-stationarity as a key challenge, and proposes synthetic tasks to improve performance across ancestry groups.

Bridging Ancestry Gaps in Genomic Risk Prediction with Tabular Foundation Models

This paper improves genomic risk prediction across ancestry groups using instruction-tuned tabular foundation models, showing better performance in under-sampled populations.

Integrative Transfer Network: Deep Transfer Learning Across Populations and Prediction Targets

The paper introduces ITN, a deep learning model that improves subgroup-specific prediction by integrating multiple outcomes and subgroups in biomedical datasets.

Equitable Health Intelligence: An Open Benchmark of Multi-Population Machine Learning for Omics-Based Cancer Prognosis

The paper introduces EHI, an open benchmark for multi-population ML in cancer prognosis, highlighting performance disparities and mitigation strategies across different populations.

Lung Adenocarcinoma Just Desserts: An Expanding Pie of Activating Oncogenes or a Layer Cake of Integrated Alterations

The paper identifies an EGFR mutation signature (mSig) that captures transcriptional features of EGFR activation in LUAD, revealing subtype-dependent oncogene activity and potential therapeutic targets.

Data Representation Bias and Conditional Distribution Shift Drive Predictive Performance Disparities in Multi-Population Machine Learning

This study evaluates how data representation bias and distribution shifts affect multi-population machine learning performance using synthetic genotype-phenotype datasets across five continental populations.

MARVEL: Microenvironment Annotation by Supervised Graph Contrastive Learning

MARVEL improves microenvironment annotation by using supervised graph contrastive learning to embed spatial cell relationships into a continuous space, outperforming existing methods in three tasks.

Deep transfer learning provides a Pareto improvement for multi-ancestral clinico-genomic prediction of diseases

This paper uses transfer learning to improve disease prediction accuracy for underrepresented populations without harming accuracy for others.

Scalable Condition-relevant Cell Niche Analysis of Spatial Omics Data with Taichi

Taichi identifies condition-relevant cell niches in spatial omics data by analyzing intra-slice heterogeneity across multi-slice and multi-condition datasets.

Clinical Time-to-Event Prediction Enhanced by Incorporating Compatible Related Outcomes

This paper improves time-to-event prediction by integrating compatible related clinical outcomes, enhancing performance without requiring auxiliary data during prediction.

Gene-by-environmental modulation of longevity and weight gain in the murine BXD family

The paper examines how diet and genetics interact to affect longevity and weight gain in mice, finding that genetic differences significantly influence the impact of high-fat diets on lifespan.

Deep transfer learning for reducing health care disparities arising from biomedical data inequality

This study shows transfer learning can reduce health care disparities by improving model performance for underrepresented racial groups in cancer omics data.

Differential gene expression analysis reveals novel genes and pathways in pediatric septic shock patients

The paper identifies 54 differentially expressed genes in pediatric septic shock patients and links them to immune response and chemokine pathways, distinguishing survivors from non-survivors.