Machine learning analysis of TCGA cancer data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Machine learning (ML) is gaining traction in analyzing complex biological problems.
• Large-scale initiatives like The Cancer Genome Atlas (TCGA) provide omic data for ML model training.
• Standard analytical approaches are often insufficient for intricate biological data.

Purpose of the Study:

• To review and classify existing literature on ML applications using TCGA data.
• To identify key trends, popular algorithms, and biological problems addressed.
• To provide a foundation for future research in ML for cancer genomics.

Main Methods:

• Systematic review of over 100 research papers utilizing TCGA data for ML model training.
• Classification of studies based on tumor type, ML algorithm employed, and the biological question investigated.
• Analysis of data types used, with a focus on gene expression data.

Main Results:

• Random Forest and Support Vector Machines are dominant ML algorithms, with a rise in deep neural networks.
• Integrative multi-omic data analysis models are increasingly utilized.
• Gene expression data is the most frequently used data type for training ML models.

Conclusions:

• ML is a powerful tool for dissecting cancer genomics, leveraging TCGA data.
• Specific tumor types (e.g., BRCA) and biological problems (e.g., survival prediction in GBM) show distinct research trends.
• This review consolidates the state-of-the-art, guiding future ML-driven cancer research.