Evaluating the Predictability of Cancer Types from 536 Somatic Mutations: A New Dataset

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning in Oncology

Background:

• Somatic mutations in cancer genes are crucial for understanding cancer development and progression.
• The Cancer Genome Atlas (TCGA) project has generated extensive genomic data for various cancer types.
• Developing computational tools to analyze mutation data is essential for advancing cancer research.

Purpose of the Study:

• To introduce a novel dataset comprising somatic mutation states for 536 Cancer Gene Census (CGC) genes.
• To evaluate the efficacy of machine learning models in classifying cancer types based on somatic mutations.
• To provide a valuable resource for TCGA data research, including survival and image analysis.

Main Methods:

• Compilation of somatic mutation data from TCGA projects for 536 CGC genes.
• Application of machine learning algorithms: k-Nearest Neighbors, Decision Tree, Random Forest, and Artificial Neural Networks (ANNs).
• Comparative analysis of model performance using accuracy, precision, recall, and F1-score metrics.

Main Results:

• Artificial Neural Networks (ANNs) demonstrated superior performance, achieving an F1-score of 0.36 and 40% overall classification accuracy.
• The ANN model's accuracy (40%) significantly surpasses random guessing (3%), with an average specificity of 98%.
• High precision (>0.7) was achieved by the ANN for 5 out of 33 cancer types, indicating potential for specific classifications.

Conclusions:

• Somatic mutation data from TCGA, when analyzed with machine learning, holds promise for cancer type classification.
• The developed dataset serves as a foundational resource for further investigations in cancer genomics and personalized medicine.
• The dataset is publicly available, facilitating broader research applications in survival analysis and histopathology image analysis.