Gene characteristics predicting missense, nonsense and frameshift mutations in tumor samples

Area of Science:

• Genomics
• Cancer Biology
• Bioinformatics

Background:

• Driver mutations confer a selective advantage in tumors, leading to higher frequencies than neutral mutations.
• Mutation frequency alone is insufficient for cancer gene identification due to varying gene mutability influenced by factors like size and nucleotide composition.

Purpose of the Study:

• To identify specific gene characteristics associated with the frequency of somatic mutations in tumor samples.
• To develop a predictive model for identifying cancer-associated genes.

Main Methods:

• Utilized somatic mutation data from the Catalog of Somatic Mutations in Cancer (COSMIC).
• Employed stepwise multiple linear regression with 11 gene characteristics (size, nucleotide composition, expression, replication time, conservation) as predictors.
• Modeled missense, nonsense, and frameshift mutations separately due to differing characteristic associations.

Main Results:

• Gene characteristics explained significant variation: 88% for missense, 40% for nonsense, and 23% for frameshift mutations.
• Comparison of observed and expected mutation counts identified known driver genes and 111 novel candidate driver genes.
• Incorporating silent mutation counts improved prediction accuracy beyond existing tools like MutsigCV.

Conclusions:

• Comparing observed and predicted mutation numbers effectively identifies known and novel cancer-associated genes.
• The developed method significantly enhances the accuracy of cancer gene prediction tools.