Machine Learning-Based Analysis of Glioma Grades Reveals Co-Enrichment
Mateusz Garbulowski1,2, Karolina Smolinska1, Uğur Çabuk1,3,4
1Department of Cell and Molecular Biology, Uppsala University, 752 37 Uppsala, Sweden.
Cancers
|February 25, 2022
Summary
We developed an interpretable machine learning method to analyze The Cancer Genome Atlas (TCGA) glioma data, correcting batch effects and revealing co-predictive mechanisms for improved glioma grading and targeted treatment strategies.
Area of Science:
- Neuro-oncology
- Computational Biology
- Genomics
Background:
- Gliomas are primary brain tumors with complex grading systems.
- The Cancer Genome Atlas (TCGA) provides extensive transcriptomics data for gliomas.
- Batch effects in large datasets like TCGA can confound analyses.
Purpose of the Study:
- To apply interpretable machine learning to TCGA glioma data.
- To identify co-predictive mechanisms among biomarkers for glioma grading.
- To improve the accuracy and reliability of glioma subtyping and treatment.
Main Methods:
- Corrected batch effects in TCGA glioma transcriptomics data.
- Utilized interpretable machine learning on gene set enrichment scores.
- Employed rule-based classifiers to identify co-enrichment networks.
- Validated findings using external glioma cohorts.
Main Results:
- Successfully corrected batch effects in TCGA glioma data.
- Discovered networks of co-enrichment related to glioma grades.
- Demonstrated the predictive power of interpretable models.
- Validated the identified co-enrichment patterns in external datasets.
Conclusions:
- Corrected TCGA glioma datasets enhance the reliability of future studies.
- Co-enrichment networks offer insights into glioma progression.
- This approach supports the development of targeted glioma treatments.
