Wei Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Wei Zhang has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
- Steven Chen
Department of Public Health Sciences, University of Miami Miller School of Medicine
4 shared publications Hanchen Huang
University of Miami Miller School of Medicine Division of Biostatistics and Bioinformatics, Department of Public Health Sciences Miami, FL 33136
3 shared publicationsLily Wang
University of Miami Miller School of Medicine Division of Biostatistics and Bioinformatics, Department of Public Health Sciences Miami, FL 33136
2 shared publications- Lily Wang
Division of Biostatistics, Department of Public Health Sciences, University of Miami, Miller School of Medicine, Miami, USA
2 shared publications Achintya Varma
John P. Hussman Institute for Human Genomics, University of Miami Miller School of Medicine, Miami, USA
1 shared publicationBrian D Lehmann
Vanderbilt University Medical Center Division of Hematology and Oncology, Department of Medicine Nashville, TN 37232
1 shared publication
Latest Publications
- Open Access
- 9 December 2025
- Published by Oxford University Press (OUP) in GigaScience
- Vol. 15
An integrative multiomics random forest framework for robust biomarker discovery
Background: High-throughput technologies now produce a wide array of omics data, from genomic and transcriptomic profiles to epigenomic and proteomic measurements. Integrating multiple omics layers measured on the same samples can reveal cross-layer molecular hubs that single-layer analyses miss. However, many existing integrative methods rely on linear assumptions or univariate feature importance, limiting their ability to capture nonlinear and interaction-driven dependencies across data modalities. Results: We present an unsupervised, multivariate random forest (MRF) framework with an inverse minimal depth (IMD) importance to prioritize shared biomarkers across omics. In each forest, one layer serves as a multivariate response and the other as predictors; IMD summarizes how early a predictor (or response maximal splitting response variable) appears across trees, yielding interpretable, cross-layer feature rankings. We provide two IMD-based selection strategies and introduce an optional IMD power transform to enhance sensitivity to interaction signals. In extensive simulations spanning linear, nonlinear, and interaction regimes, our method matches sparse partial least squares/canonical correlation analysis under linear settings and outperforms them as nonlinearity increases, while adapted univariate ensemble learners (random forest, gradient boosting machine, XGBoost) underperform in the multivariate, unsupervised context. Applied to breast invasive carcinoma and colon adenocarcinoma in The Cancer Genome Atlas (TCGA), MRF-IMD identifies genes, CpGs, and microRNAs enriched for cancer-relevant pathways and yields more robust survival stratification than linear integrators with matched model sizes. In a TCGA pan-cancer analysis, MRF-IMD features achieve a higher Adjusted Rand Index than alternatives and recover coherent tumor-type clusters; in the Alzheimer’s Disease Neuroimaging Initiative (ADNI), the integrative signature improves dementia progression stratification over a published methylation risk score. Conclusions: MRF-IMD provides a scalable and interpretable framework for multiomics integration that reliably identifies cross-layer biomarkers when nonlinear and interaction-driven dependencies are present. This approach advances robust biomarker discovery beyond the limits of linear integrative methods.
- 9 December 2025
- Published by Oxford University Press (OUP) in GigaScience
- Vol. 15
Open AccessAn integrative multiomics random forest framework for robust biomarker discovery
Background: High-throughput technologies now produce a wide array of omics data, from genomic and transcriptomic profiles to epigenomic and proteomic measurements. Integrating multiple omics layers measured on the same samples can reveal cross-layer molecular hubs that single-layer analyses miss. However, many existing integrative methods rely on linear assumptions or univariate feature importance, limiting their ability to capture nonlinear and interaction-driven dependencies across data modalities. Results: We present an unsupervised, multivariate random forest (MRF) framework with an inverse minimal depth (IMD) importance to prioritize shared biomarkers across omics. In each forest, one layer serves as a multivariate response and the other as predictors; IMD summarizes how early a predictor (or response maximal splitting response variable) appears across trees, yielding interpretable, cross-layer feature rankings. We provide two IMD-based selection strategies and introduce an optional IMD power transform to enhance sensitivity to interaction signals. In extensive simulations spanning linear, nonlinear, and interaction regimes, our method matches sparse partial least squares/canonical correlation analysis under linear settings and outperforms them as nonlinearity increases, while adapted univariate ensemble learners (random forest, gradient boosting machine, XGBoost) underperform in the multivariate, unsupervised context. Applied to breast invasive carcinoma and colon adenocarcinoma in The Cancer Genome Atlas (TCGA), MRF-IMD identifies genes, CpGs, and microRNAs enriched for cancer-relevant pathways and yields more robust survival stratification than linear integrators with matched model sizes. In a TCGA pan-cancer analysis, MRF-IMD features achieve a higher Adjusted Rand Index than alternatives and recover coherent tumor-type clusters; in the Alzheimer’s Disease Neuroimaging Initiative (ADNI), the integrative signature improves dementia progression stratification over a published methylation risk score. Conclusions: MRF-IMD provides a scalable and interpretable framework for multiomics integration that reliably identifies cross-layer biomarkers when nonlinear and interaction-driven dependencies are present. This approach advances robust biomarker discovery beyond the limits of linear integrative methods.
- Open Access
An Integrative Multi-Omics Random Forest Framework for Robust Biomarker Discovery
High-throughput technologies now produce a wide array of omics data, from genomic and transcriptomic profiles to epigenomic and proteomic measurements. Integrating these diverse data types can yield deeper insights into the biological mechanisms driving complex traits and diseases. Yet, extracting key shared biomarkers from multiple data layers remains a major challenge. We present a multivariate random forest (MRF)–based framework enhanced by a novel inverse minimal depth (IMD) metric for integrative variable selection. By assigning response variables to tree nodes and employing IMD to rank predictors, our approach efficiently identifies essential features across different omics types, even when confronted with high-dimensionality and noise. Through extensive simulations and analyses of multi-omics datasets from The Cancer Genome Atlas, we demonstrate that our method outperforms established integrative techniques in uncovering biologically meaningful biomarkers and pathways. Our findings show that selected biomarkers not only correlate with known regulatory and signaling networks but can also stratify patient subgroups with distinct clinical outcomes. The method’s scalable, interpretable, and user-friendly implementation ensures broad applicability to a range of research questions. This MRF-based framework advances robust biomarker discovery and integrative multi-omics analyses, accelerating the translation of complex molecular data into tangible biological and clinical insights.
Open AccessAn Integrative Multi-Omics Random Forest Framework for Robust Biomarker Discovery
High-throughput technologies now produce a wide array of omics data, from genomic and transcriptomic profiles to epigenomic and proteomic measurements. Integrating these diverse data types can yield deeper insights into the biological mechanisms driving complex traits and diseases. Yet, extracting key shared biomarkers from multiple data layers remains a major challenge. We present a multivariate random forest (MRF)–based framework enhanced by a novel inverse minimal depth (IMD) metric for integrative variable selection. By assigning response variables to tree nodes and employing IMD to rank predictors, our approach efficiently identifies essential features across different omics types, even when confronted with high-dimensionality and noise. Through extensive simulations and analyses of multi-omics datasets from The Cancer Genome Atlas, we demonstrate that our method outperforms established integrative techniques in uncovering biologically meaningful biomarkers and pathways. Our findings show that selected biomarkers not only correlate with known regulatory and signaling networks but can also stratify patient subgroups with distinct clinical outcomes. The method’s scalable, interpretable, and user-friendly implementation ensures broad applicability to a range of research questions. This MRF-based framework advances robust biomarker discovery and integrative multi-omics analyses, accelerating the translation of complex molecular data into tangible biological and clinical insights.
- Open Access
- 5 January 2025
- Published by Elsevier in Translational Oncology
- Vol. 52, 102238
Enhancing chemotherapy response prediction via matched colorectal tumor-organoid gene expression analysis and network-based biomarker selection
- 5 January 2025
- Published by Elsevier in Translational Oncology
- Vol. 52, 102238
Open AccessEnhancing chemotherapy response prediction via matched colorectal tumor-organoid gene expression analysis and network-based biomarker selection
- Open Access
Transcriptome Meta-Analysis of Triple-Negative Breast Cancer Response to Neoadjuvant Chemotherapy
Triple-negative breast cancer (TNBC) is a heterogeneous disease with varying responses to neoadjuvant chemotherapy (NAC). The identification of biomarkers to predict NAC response and inform personalized treatment strategies is essential. In this study, we conducted large-scale gene expression meta-analyses to identify genes associated with NAC response and survival outcomes. The results showed that immune, cell cycle/mitotic, and RNA splicing-related pathways were significantly associated with favorable clinical outcomes. Furthermore, we integrated and divided the gene association results from NAC response and survival outcomes into four quadrants, which provided more insights into potential NAC response mechanisms and biomarker discovery.
Open AccessTranscriptome Meta-Analysis of Triple-Negative Breast Cancer Response to Neoadjuvant Chemotherapy
Triple-negative breast cancer (TNBC) is a heterogeneous disease with varying responses to neoadjuvant chemotherapy (NAC). The identification of biomarkers to predict NAC response and inform personalized treatment strategies is essential. In this study, we conducted large-scale gene expression meta-analyses to identify genes associated with NAC response and survival outcomes. The results showed that immune, cell cycle/mitotic, and RNA splicing-related pathways were significantly associated with favorable clinical outcomes. Furthermore, we integrated and divided the gene association results from NAC response and survival outcomes into four quadrants, which provided more insights into potential NAC response mechanisms and biomarker discovery.
- Open Access
- 15 September 2022
- Published by Springer Nature in Alzheimer's Research & Therapy
- Vol. 14 (1) , 1-21
Distinct sex-specific DNA methylation differences in Alzheimer’s disease
Background: Sex is increasingly recognized as a significant factor contributing to the biological and clinical heterogeneity in AD. There is also growing evidence for the prominent role of DNA methylation (DNAm) in Alzheimer’s disease (AD). Methods: We studied sex-specific DNA methylation differences in the blood samples of AD subjects compared to cognitively normal subjects, by performing sex-specific meta-analyses of two large blood-based epigenome-wide association studies (ADNI and AIBL), which included DNA methylation data for a total of 1284 whole blood samples (632 females and 652 males). Within each dataset, we used two complementary analytical strategies, a sex-stratified analysis that examined methylation to AD associations in male and female samples separately, and a methylation-by-sex interaction analysis that compared the magnitude of these associations between different sexes. After adjusting for age, estimated immune cell type proportions, batch effects, and correcting for inflation, the inverse-variance fixed-effects meta-analysis model was used to identify the most consistent DNAm differences across datasets. In addition, we also evaluated the performance of the sex-specific methylation-based risk prediction models for AD diagnosis using an independent external dataset. Results: In the sex-stratified analysis, we identified 2 CpGs, mapped to the PRRC2A and RPS8 genes, significantly associated with AD in females at a 5% false discovery rate, and an additional 25 significant CpGs (21 in females, 4 in males) at P-value < 1×10−5. In methylation-by-sex interaction analysis, we identified 5 significant CpGs at P-value < 10−5. Out-of-sample validations using the AddNeuroMed dataset showed in females, the best logistic prediction model included age, estimated immune cell-type proportions, and methylation risk scores (MRS) computed from 9 of the 23 CpGs identified in AD vs. CN analysis that are also available in AddNeuroMed dataset (AUC = 0.74, 95% CI: 0.65–0.83). In males, the best logistic prediction model included only age and MRS computed from 2 of the 5 CpGs identified in methylation-by-sex interaction analysis that are also available in the AddNeuroMed dataset (AUC = 0.70, 95% CI: 0.56–0.82). Conclusions: Overall, our results show that the DNA methylation differences in AD are largely distinct between males and females. Our best-performing sex-specific methylation-based prediction model in females performed better than that for males and additionally included estimated cell-type proportions. The significant discriminatory classification of AD samples with our methylation-based prediction models demonstrates that sex-specific DNA methylation could be a predictive biomarker for AD. As sex is a strong factor underlying phenotypic variability in AD, the results of our study are particularly relevant for a better understanding of the epigenetic architecture that underlie AD and for promoting precision medicine in AD.
- 15 September 2022
- Published by Springer Nature in Alzheimer's Research & Therapy
- Vol. 14 (1) , 1-21
Open AccessDistinct sex-specific DNA methylation differences in Alzheimer’s disease
Background: Sex is increasingly recognized as a significant factor contributing to the biological and clinical heterogeneity in AD. There is also growing evidence for the prominent role of DNA methylation (DNAm) in Alzheimer’s disease (AD). Methods: We studied sex-specific DNA methylation differences in the blood samples of AD subjects compared to cognitively normal subjects, by performing sex-specific meta-analyses of two large blood-based epigenome-wide association studies (ADNI and AIBL), which included DNA methylation data for a total of 1284 whole blood samples (632 females and 652 males). Within each dataset, we used two complementary analytical strategies, a sex-stratified analysis that examined methylation to AD associations in male and female samples separately, and a methylation-by-sex interaction analysis that compared the magnitude of these associations between different sexes. After adjusting for age, estimated immune cell type proportions, batch effects, and correcting for inflation, the inverse-variance fixed-effects meta-analysis model was used to identify the most consistent DNAm differences across datasets. In addition, we also evaluated the performance of the sex-specific methylation-based risk prediction models for AD diagnosis using an independent external dataset. Results: In the sex-stratified analysis, we identified 2 CpGs, mapped to the PRRC2A and RPS8 genes, significantly associated with AD in females at a 5% false discovery rate, and an additional 25 significant CpGs (21 in females, 4 in males) at P-value < 1×10−5. In methylation-by-sex interaction analysis, we identified 5 significant CpGs at P-value < 10−5. Out-of-sample validations using the AddNeuroMed dataset showed in females, the best logistic prediction model included age, estimated immune cell-type proportions, and methylation risk scores (MRS) computed from 9 of the 23 CpGs identified in AD vs. CN analysis that are also available in AddNeuroMed dataset (AUC = 0.74, 95% CI: 0.65–0.83). In males, the best logistic prediction model included only age and MRS computed from 2 of the 5 CpGs identified in methylation-by-sex interaction analysis that are also available in the AddNeuroMed dataset (AUC = 0.70, 95% CI: 0.56–0.82). Conclusions: Overall, our results show that the DNA methylation differences in AD are largely distinct between males and females. Our best-performing sex-specific methylation-based prediction model in females performed better than that for males and additionally included estimated cell-type proportions. The significant discriminatory classification of AD samples with our methylation-based prediction models demonstrates that sex-specific DNA methylation could be a predictive biomarker for AD. As sex is a strong factor underlying phenotypic variability in AD, the results of our study are particularly relevant for a better understanding of the epigenetic architecture that underlie AD and for promoting precision medicine in AD.