Using The Cancer Genome Atlas from cBioPortal to Develop Genomic Datasets for Machine Learning Assisted Cancer...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Predicting the pathogenicity of genetic mutations is vital for disease understanding, particularly in cancer.
• Polymorphism Phenotyping (PolyPhen) and Sorting Intolerant From Tolerant (SIFT) are essential tools for assessing mutation impact.
• A lack of readily available genomic datasets hinders the development of predictive models for harmful mutations.

Purpose of the Study:

• To develop and evaluate genomic and non-genomic datasets for predicting PolyPhen and SIFT scores.
• To apply machine learning models for classifying potentially harmful genetic mutations.
• To assess the performance of different machine learning models in mutation prediction.

Main Methods:

• Utilized The Cancer Genome Atlas (TCGA) data from cBioPortal to construct genomic and non-genomic datasets.
• Implemented and compared three machine learning classification models: Random Forest (RF), Extreme Gradient Boosting (XGBoost), and an ensemble RF-XGBoost model.
• Evaluated model performance based on prediction accuracy for PolyPhen and SIFT scores.

Main Results:

• Genomic data demonstrated superior performance in predicting PolyPhen and SIFT scores compared to non-genomic data.
• The ensemble RF-XGBoost model achieved the highest prediction accuracies on genomic data: 88.43% for PolyPhen and 95.13% for SIFT.
• Machine learning models, particularly when trained on genomic data, show significant potential for accurate mutation impact prediction.

Conclusions:

• Genomic datasets are more effective for building accurate predictive models of genetic mutation impact.
• The ensemble RF-XGBoost model offers a powerful approach for identifying potentially harmful mutations.
• Artificial intelligence holds significant promise for advancing genetic mutation analysis in the context of disease research and clinical applications.