Related Experiment Video
Updated: Jun 10, 2026
08:50
Predictive Immune Modeling of Solid Tumors
Published on: February 25, 2020
In silico estimates of tissue components in surgical samples based on expression profiling data.
Yipeng Wang1, Xiao-Qin Xia, Zhenyu Jia
1Department of Pathology and Laboratory Medicine, Vaccine Research Institute of San Diego, University of California, Irvine, Irvine, California 92067, USA.
Cancer Research
|July 29, 2010
Summary
Gene expression profiling can be confounded by varying cell types in tissue samples. This study developed a computational method to predict tissue components from expression data, improving sample analysis for disease research.
Area of Science:
- Computational biology
- Bioinformatics
- Cancer research
Background:
- Gene expression profiling studies often use tissue samples with diverse cell compositions.
- Variations in cell types can complicate the correlation of gene expression with clinical parameters.
- Accurate estimation of tissue components is crucial for reliable data analysis.
Purpose of the Study:
- To develop and validate multivariate linear regression models for in silico prediction of major tissue components (tumor and stroma) from gene expression data.
- To assess the performance of these models using independent datasets and literature-derived samples.
- To investigate the impact of tissue composition on gene expression correlations with clinical parameters, specifically prostate cancer recurrence.
Main Methods:
- Utilized four large gene expression microarray datasets from prostate cancer patients with pathologist-estimated tissue components.
- Employed multivariate linear regression models and ten-fold cross-validation for in silico prediction.
- Validated models across independent datasets and applied them to literature-derived "tumor-enriched" samples.
- Analyzed the correlation between predicted tissue percentages and clinical parameters like cancer recurrence.
Main Results:
- In silico predictions showed average differences of 8-17% compared to pathologist estimates within and across datasets.
- Nearly a quarter of "tumor-enriched" samples were predicted to contain 30% or less tumor cells.
- A significant difference in predicted tumor content was observed between recurrent and non-recurrent cancer patients.
- Genes correlating with recurrence also correlated with predicted tissue percentage, highlighting the confounding effect of stromal content.
Conclusions:
- Multivariate linear regression models can accurately predict tissue components from gene expression data.
- In silico prediction of tissue composition is essential for accurate interpretation of gene expression profiling studies, especially in cancer research.
- The developed models and the 'CellPred' web service can help researchers triage samples and account for tissue heterogeneity in their analyses to avoid spurious correlations.
