XLXiaomei Li

Xiaomei Li

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Xiaomei Li has a total of 98 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    TabSurv: Tabular foundation model for breast cancer prognosis using gene expression data

    by Tuyen Vu, Ha Xuan Tran, Xiaomei Li, Sindy Pinero, Lin Liu, Jiuyong Li, Jia Tina Du, Thuc Duy Le

  • Open Access

    TACO: TabPFN augmented causal outcomes for early detection of long COVID

    by Sindy Pinero, Ngoc Duong, Xiaomei Li, Lin Liu, Jiuyong Li, Sang Hong Lee, Marnie Winter, Thin Nguyen, Junpeng Zhang, Thuc Duy Le

    Long COVID, or post-acute sequelae of COVID-19 (PASC), affects 10% to 40% of COVID-19 survivors worldwide, manifesting as persistent symptoms that last months after initial infection, including fatigue, cognitive impairment, and organ dysfunction. The heterogeneous nature of Long COVID and its delayed onset creates a critical window in which at-risk patients remain unidentified during the presymptomatic phase, when interventions could be most beneficial. Current prediction methods rely on clinical symptoms that manifest too late for early intervention or use molecular approaches based on statistical associations that cannot distinguish disease drivers from correlational noise. We present TACO (TabPFN-augmented causal outcomes), a novel method for the early detection of Long COVID using gene expression data. TACO first applies differential causal effect (DCE) analysis to identify genes with putative regulatory causal relationships to the pathogenesis of Long COVID, rather than simple statistical associations. It then integrates these mechanistic features with the tabular prior-data fitted network (TabPFN), a foundation model that requires no hyperparameter tuning, alongside complementary classifiers in an ensemble approach. Experimental results show that TACO performs competitively with existing machine learning methods in predicting Long COVID from gene expression data from patients with early-stage COVID-19, while offering a causally grounded and interpretable feature selection. An ablation study confirms the positive contribution of TabPFN to the ensemble, and a strict nested cross-validation protocol, in which DCE feature selection is performed exclusively within each training fold, confirms that the results are robust to data leakage. The nested CV analysis further identifies a minimal reproducible causal gene signature with potential utility for targeted clinical sequencing during acute hospitalization for COVID-19. In addition, the causal genes identified by TACO are highly relevant to Long COVID-related functions and pathways, as confirmed by the literature and subsequent enrichment analyses, providing insights into disease mechanisms and potential treatments.

  • Open Access

    SPLIT: Safety Prioritization for Long COVID Drug Repurposing via a Causal Integrated Targeting Framework

    by Sindy Pinero, Xiaomei Li, Lin Liu, Jiuyong Li, Sang Hong Lee, Thuc Duy Le

    SUMMARY: Long COVID affects millions of people worldwide, yet no disease-modifying treatment has been approved, and existing interventions have shown only modest and inconsistent benefits. A key reason for this limited progress is that current computational drug repurposing pipelines do not match well with the clinical reality of Long COVID. These patients often have persistent, multi-systemic symptoms and may already be taking multiple medications, making treatment safety a primary concern. However, most repurposing workflows still treat safety as a downstream filter and rely on disease-associated targets rather than causal drivers. They also assume that the findings of one analysis would generalize across the diverse presentations of Long COVID. We introduce SPLIT, a safety-first repurposing framework that addresses these limitations. SPLIT prioritizes safety at the start of the candidate evaluation, integrates complementary causal inference strategies to identify likely driver genes, and uses a counterfactual substitution design to compare drugs within specific cohort contexts. When applied to cognitive and respiratory Long COVID cohorts, SPLIT revealed three main findings. First, drugs with similar predicted efficacy could have very different predicted safety profiles. Second, the drugs flagged as unfavorable were often different between the two cohorts, showing that drug prioritization is phenotype-specific. Third, SPLIT flagged 18 drugs currently under active investigation in Long COVID trials as having unfavorable predicted profiles. SPLIT provides a practical framework to identify safer, more context-appropriate candidates earlier in the process, supporting more targeted and better-tolerated treatment strategies for Long COVID.

  • Omics-based computational approaches for biomarker identification, prediction, and treatment of Long COVID

    by Sindy Pinero, Xiaomei Li, Junpeng Zhang, Marnie Winter, Sang Hong Lee, Thin Nguyen, Lin Liu, Jiuyong Li, Thuc Duy Le

    Long COVID, or post-acute sequelae of COVID-19 (PASC), is a major global health problem, with cumulative estimates suggesting that around 400 million people worldwide have been affected. It is characterized by persistent or new symptoms such as fatigue, cognitive impairment, and breathlessness lasting beyond four weeks after acute infection. Diverse clinical manifestations, chronic course, and incompletely understood pathophysiology-including hypotheses involving viral persistence, immune dysregulation, autoimmunity, endothelial dysfunction, and metabolic reprogramming-impede the development of diagnostic criteria, biomarkers, and targeted therapies. We conducted a critical review of 101 Long COVID omics studies, focusing on the computational methods used and their methodological quality. Using standardized criteria, we evaluated study design, statistical rigor, reproducibility, and clinical relevance across genomics, epigenomics, transcriptomics, proteomics, metabolomics, and multiomics integration, and mapped these findings onto regulatory and translational frameworks. Despite substantial methodological heterogeneity, convergent biological signals emerged. Genomic studies implicate risk loci in immune and cardiopulmonary pathways. Epigenomic analyses identify differentially methylated regions in immune and circadian genes. Transcriptomic studies reveal persistent dysregulation of innate immune and coagulation pathways, as well as reproducible molecular endotypes. Proteomic studies consistently show abnormalities in the complement cascade and coagulation, with a small panel of complement proteins showing highly reproducible changes across independent cohorts. Metabolomic studies demonstrate sustained mitochondrial dysfunction and altered cellular bioenergetics for up to two years after infection. Multiomics integration supports at least two major endotypes, characterized by predominant inflammatory versus metabolic dysregulation, and provides a basis for patient stratification and computational treatment discovery. Machine learning models frequently achieve high classification performance, but are rarely externally validated. Critical limitations restrict clinical translation. Most studies are underpowered relative to analytical complexity, use heterogeneous case definitions and controls, and report platform-specific signatures with limited overlap. External validation, preregistered analysis plans, and regulatory-aligned assay development are uncommon. To date, no regulatory-approved diagnostic assay or evidence-based therapeutic intervention has directly emerged from these computational findings. Future progress requires harmonized phenotyping protocols, adequately powered longitudinal cohorts with external validation, integration of spatial omics and explainable artificial intelligence, and early engagement with regulatory and health-technology assessment pathways. This review provides a critical assessment and a translational roadmap, outlining how methodologically robust computational omics can be advanced toward clinically actionable tools for Long COVID.


  • Open Access

    Tabular Foundation Model for Breast Cancer Prognosis using Gene Expression Data

    by Tuyen Vu, Ha X. Tran, Xiaomei Li, Lin Liu, Jiuyong Li, Sindy Pinero, Jia Tina Du, Thuc D. Le

    Accurate and robust survival prediction is essential for personalised breast cancer prognosis and treatment decision-making. However, existing machine learning–based survival models often lack stability under cohort heterogeneity and distribution shift, while tabular foundation models typically do not support censored time-to-event data. This study aims to develop a foundation-model-based survival framework that explicitly handles right censoring and enables stable deployment across heterogeneous cohorts without task-specific retraining. We propose TabSurv, a foundation-model-based framework for survival analysis that adapts pretrained tabular foundation models to censored time-to-event prediction. Built on TabPFN, TabSurv operates via in-context learning without gradient-based training or hyper-parameter tuning. To address censoring, TabSurv introduces a two-stage, censoring-aware strategy. In the first stage, the model learns survival time relationships from uncensored observations and imputes survival outcomes for censored patients under a non-informative censoring assumption. In the second stage, imputed and observed outcomes are combined to reconstruct the learning context for final inference, reformulating survival analysis as a regression task. For treatment recommendation, TabSurv estimates individualised counterfactual survival risks using gene expression features restricted to biologically validated cancer driver genes. TabSurv was evaluated on twelve breast cancer gene expression cohorts under in-distribution and out-of-distribution settings. It achieved competitive or superior prognostic performance compared with seven established survival models and demonstrated the highest predictive stability under cohort heterogeneity. In treatment recommendation experiments on METABRIC, patients whose treatments aligned with TabSurv’s recommendations exhibited significantly improved long-term survival. TabSurv provides an efficient and stable foundation-model-based approach to censored survival analysis, supporting robust prognosis and personalised treatment recommendation in heterogeneous breast cancer cohorts.

  • Open Access

    TACO: TabPFN Augmented Causal Outcomes for Early Detection of Long COVID

    by Sindy Piñero, Xiaomei Li, Lin Liu, Jiuyong Li, Sang Hong Lee, Marnie Winter, Thin Nguyen, Junpeng Zhang, Thuc Duy Le

    Long COVID affects 10-40% of COVID-19 survivors, yet early detection remains challenging. We present TACO (TabPFN Augmented Causal Outcomes), a framework that uniquely combines causal inference with foundation models for presymptomatic Long COVID detection. TACO employs Differential Causal Effect (DCE) analysis to identify causally relevant genes, then utilizes TabPFN, a foundation model that does not require hyperparameter adjustment, to achieve consistent performance. In comprehensive benchmarking, TACO achieved superior precision using 18% fewer features than conventional approaches. Critically, TACO maintains consistent performance without any hyperparameter optimization, while benchmark models show variable results depending on the tuning. The causal genes of the framework provide biological interpretability, with 23.6% validated in the Long COVID literature (4.72-fold enrichment, p = 5.04 × 10 −39 ), including regulators of viral entry ( AR, TMPRSS2 ), immune response ( TP53, CDKN1A ), and tissue remodeling ( SMAD2/3 ). By prioritizing causal mechanisms over statistical associations and eliminating the need for hyperparameter search, TACO offers a practical, interpretable solution for clinical deployment, transforming Long COVID management from reactive diagnosis to proactive prevention.


  • Open Access

    A conceptual framework for human–AI collaborative genome annotation

    by Xiaomei Li, Alex Whan, Meredith McNeil, David Starns, Jessica Irons, Samuel C Andrew, Rad Suchecki

    Genome annotation is essential for understanding the functional elements within genomes. While automated methods are indispensable for processing large-scale genomic data, they often face challenges in accurately predicting gene structures and functions. Consequently, manual curation by domain experts remains crucial for validating and refining these predictions. These combined outcomes from automated tools and manual curation highlight the importance of integrating human expertise with artificial intelligence (AI) capabilities to improve both the accuracy and efficiency of genome annotation. However, the manual curation process is inherently labor-intensive and time-consuming, making it difficult to scale for large datasets. To address these challenges, we propose a conceptual framework, Human-AI Collaborative Genome Annotation (HAICoGA), that leverages the synergistic partnership between humans and AI to enhance human capabilities and accelerate the genome annotation process. Additionally, we explore the potential of integrating large language models into this framework to support and augment specific tasks. Finally, we discuss emerging challenges and outline open research questions to guide further exploration in this area.


  • Open Access

    A Novel Method for Trust-Sensitive Design: Applying Conjoint Analysis to Machine-Assisted Genome Annotation

    by Melanie McGrath, Patrick Cooper, Andreas Duenser, Jessica Irons, Xiaomei Li, Rad Suchecki

  • Open Access

    Omics-Based Computational Approaches for Biomarker Identification, Prediction, and Treatment of Long COVID

    by Sindy Licette Pinero, Xiaomei Li, Junpeng Zhang, Marnie Winter, Sang Hong Lee, Thin Nguyen, Lin Liu, Jiuyong Li, Thuc Duy Le

    Long COVID, also referred to as post-acute sequelae of COVID-19 (PASC), is a substantial global health concern estimated to have affected over 145 million individuals worldwide. Characterized by persistent and new symptoms extending beyond four weeks from the initial infection—such as fatigue, breathlessness, and cognitive impairments—Long COVID poses significant challenges to healthcare systems due to its chronic nature and diverse clinical manifestations. An urgent need exists to elucidate its underlying mechanisms to facilitate the development of effective diagnostic tools and targeted treatments. This paper provides a comprehensive overview of the current datasets and computational methods to investigate the causes, risk factors, and potential treatments for Long COVID. To understand better the molecular processes driving this condition, we examine various omics data sources, including genomics, epigenomics, transcriptomics, proteomics, and metabolomics datasets. Moreover, we discuss how integrating multi-omics data and using advanced computational techniques—such as machine learning and network analysis—can enhance Long COVID diagnosis, prognosis, and therapeutic strategies. We also emphasize the importance of larger, more diverse cohorts, longitudinal research designs, and cross-cohort validations to strengthen the reliability and applicability of findings. By addressing these crucial aspects, this review aims to advance the development of effective interventions for Long COVID, ultimately improving patient outcomes and alleviating the disease burden.


  • Open Access

    GeneWhisperer: Enhancing manual genome annotation with large language models

    by Xiaomei Li, Alex P Whan, Meredith McNeil, Samuel C. Andrew, Xiang Dai, Madeline Fechner, Cécile Paris, Rad Suchecki

    Genome annotation is critical for understanding functional elements within genomes. Manual curation is a common practice for identifying the functions of genes, particularly those missed by automated annotation pipelines. However, this process is notoriously labour-intensive and time-consuming. In response to these challenges, we present GeneWhisperer, an innovative assistant system designed to facilitate the manual gene functional annotation process. Utilizing a large language model (LLM) agent, GeneWhisperer provides users access to tools appropriate for specific curation tasks in genome annotation. Featuring a conversational interface, GeneWhisperer fosters effective collaboration between human experts and AI. By synergizing the capabilities of AI with human expertise, GeneWhisperer offers a promising approach to advancing our understanding of gene functions and streamlining the gene annotation process.