Inferring tumor purity using multi-omics data based on a uniform machine learning framework MoTP.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Current tumor purity assessment algorithms are restricted to single omics data types.
• This limitation hinders comprehensive and accurate estimation of tumor purity.

Purpose of the Study:

• To develop and validate the Multi-omics Tumor Purity prediction (MoTP) algorithm for enhanced tumor purity estimation.
• To leverage multiple omics data types for improved predictive performance.

Main Methods:

• Utilized Bayesian Regularized Neural Networks for prediction.
• Trained MoTP using integrated multi-omics data (mRNA, microRNA, lncRNA, DNA methylation) from 21 TCGA cancer types.
• Validated MoTP on TCGA and external datasets, including those with noise and missing features.

Main Results:

• MoTP demonstrated excellent performance, with multi-omics integration significantly enhancing prediction accuracy over single omics data.
• The algorithm showed robustness when tested on noisy and incomplete datasets.
• Benchmark analysis confirmed MoTP's superior performance compared to established algorithms, with reduced computational requirements.

Conclusions:

• MoTP offers a powerful and efficient machine learning approach for computational tumor purity inference.
• The integration of multi-omics data provides a more accurate and reliable estimation of tumor purity.
• MoTP represents an attractive advancement for cancer research and clinical applications.