Uniformly shaped harmonization combines human transcriptomic data from different platforms while retaining their...

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Comparing RNA profiles across different platforms is challenging due to data incompatibility.
• Existing normalization tools produce dataset-specific outputs, hindering cross-platform analysis.
• A universal data format is needed for comprehensive analysis of gene expression.

Purpose of the Study:

• To evaluate the Shambhala approach for harmonizing gene expression data.
• To assess Shambhala's ability to retain biological features like fold-change expression and pathway activation.
• To compare Shambhala's performance against other transcriptomic harmonization methods.

Main Methods:

• Applied Shambhala and other harmonization methods to 6,793 cancer and 11,135 normal tissue gene expression profiles.
• Evaluated performance using twelve criteria, including biological classification, clustering, correlation, and drug activity prediction.
• Tested Shambhala-2's effectiveness in maintaining data quality for machine learning classifiers.

Main Results:

• Shambhala-2 demonstrated superior performance in harmonizing transcriptomic data.
• Achieved high correlation and linear regression coefficients between training and validation datasets.
• Showed over two times less instability in calculating drug efficiency scores compared to other methods.

Conclusions:

• Shambhala-2 effectively harmonizes RNA profiles, preserving biological signals across diverse datasets.
• The uniform output format enhances data compatibility for comparative transcriptomic studies.
• Shambhala-2 shows promise for improving disease research and cancer drug activity prediction.