Computational algorithms for in silico profiling of activating mutations in cancer

Area of Science:

• Computational Biology
• Genomics
• Protein Science

Background:

• Assessing the impact of protein mutations in human cancers is crucial for understanding disease mechanisms and developing targeted therapies.
• Existing computational methods include data-driven (evolutionary, machine learning) and structure-based approaches for predicting mutation effects.

Purpose of the Study:

• To compare the efficacy of structure-based versus data-driven computational methods for analyzing protein mutational landscapes in cancer.
• To evaluate these methods using case studies of mutations in Anaplastic Lymphoma Kinase (ALK), B-Raf (BRAF), and Human Epidermal growth factor Receptor 2 (HER2).

Main Methods:

• Comparative analysis of two distinct computational approaches: structure-based prediction and data-driven (evolutionary/machine learning) methods.
• Case studies involving kinase domain mutations in ALK (neuroblastoma), BRAF (melanoma), and HER2 (breast cancer).

Main Results:

• Structure-based methods are effective for binary classification of mutations, particularly rare ones, and predicting inhibitor interactions.
• Data-driven methods are more suitable for identifying broad mutational patterns across multiple proteins when structure-based models are computationally intensive.
• Significant differences in accuracy and confidence values were observed between the two methods.

Conclusions:

• The optimal choice between structure-based and data-driven computational methods for cancer mutation analysis is context-dependent.
• Structure-based approaches offer high precision for specific mutation effects, while data-driven methods provide broader pattern recognition capabilities.
• Future research should consider the complementary strengths of both methodologies for comprehensive cancer genomics.