Identification of potential driver mutations in glioblastoma using machine learning

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Glioblastoma (GBM) is an aggressive brain and spinal cord tumor.
• Mutations in GBM-related proteins can alter their structure and function, leading to disease.
• Distinguishing driver mutations (causative) from passenger mutations (neutral) is crucial for understanding GBM pathogenesis.

Purpose of the Study:

• To identify sequence-based features differentiating driver from passenger mutations in glioblastoma.
• To develop a machine learning model for classifying glioblastoma mutations.
• To provide a tool for prioritizing driver mutations and identifying therapeutic targets.

Main Methods:

• Collected and analyzed 9386 driver and 8728 passenger mutations in glioblastoma.
• Examined amino acid preferences and neighboring residue motifs at mutation sites.
• Computed sequence-based features including conservation scores and Position Specific Scoring Matrices (PSSM).
• Developed and validated a machine learning classifier, GBMDriver.

Main Results:

• Specific amino acids (e.g., Arg) and motifs (e.g., YP, CP, GRH) are preferred in glioblastoma driver mutations.
• Distinct amino acid preferences (Met, Ile) and motifs (SI, IQ, TVI) characterize passenger mutations.
• The GBMDriver tool achieved 73.59% accuracy and 0.82 AUC on cross-validation, and 81.99% accuracy and 0.87 AUC on a blind test set.

Conclusions:

• Sequence-based features effectively distinguish glioblastoma driver from passenger mutations.
• The developed GBMDriver tool can aid in prioritizing driver mutations for glioblastoma.
• This approach assists in identifying potential therapeutic targets for glioblastoma treatment.