TP53_PROF: a machine learning model to predict impact of missense mutations in TP53

Area of Science:

• Oncology
• Genetics
• Bioinformatics

Background:

• Precision oncology faces challenges in identifying low-frequency driver mutations.
• Accurate TP53 mutation classification is critical for both somatic and germline variants, impacting cancer predisposition like Li-Fraumeni syndrome (LFS).

Purpose of the Study:

• To develop a gene-specific machine learning model, TP53_PROF, for predicting the functional consequences of all possible TP53 missense mutations.
• To create a clinically applicable tool for accurate TP53 mutation classification.

Main Methods:

• Integrated human cell- and yeast-based functional assay scores with computational predictions.
• Trained the model using variants labeled by prevalence in four cancer genomics databases.
• Validated predictions against experimental data, population data, LFS datasets, ClinVar, and TCGA survival data.

Main Results:

• TP53_PROF achieved 96.5% prediction accuracy.
• Model predictions demonstrated high accuracy across all validation methods.
• The model effectively classifies TP53 missense mutations for clinical use.

Conclusions:

• TP53_PROF offers a validated, clinically oriented approach for TP53 mutation assessment.
• This gene-specific machine learning strategy integrates biological knowledge and data for accurate variant classification.
• The approach is poised for expansion to other critical cancer genes.