Machine learning-based models for genomic predicting neoadjuvant chemotherapeutic sensitivity in cervical cancer

Area of Science:

• Genomics and Bioinformatics
• Oncology
• Pharmacogenomics

Background:

• The PI3K/Akt pathway is implicated in platinum-based neoadjuvant chemotherapy (NACT) resistance in locally advanced cervical cancer (LACC).
• Single nucleotide polymorphisms (SNPs) represent individual genetic variations influencing treatment outcomes.
• Machine learning, specifically Random Forest (RF), offers high accuracy in predicting drug sensitivity.

Purpose of the Study:

• To apply a Random Forest (RF) model for the genomic prediction of NACT sensitivity in LACC patients.
• To identify specific genetic variations (SNPs) associated with chemoresistance.
• To explore the role of the PI3K/Akt pathway in predicting treatment response.

Main Methods:

• A cohort of 259 LACC patients was classified into effective and ineffective NACT groups.
• Genotyping of 24 SNPs across four genes (PTEN, PIK3CA, Akt1, Akt2) was performed using the Sequenom MassArray system.
• An RF model was trained using SNPs as features to predict NACT response and determine SNP importance via mean decrease in impurity.

Main Results:

• The RF model demonstrated accurate prediction of NACT response in LACC patients.
• Top predictive SNPs (rs4558508, rs1130233, rs7259541) and other significant loci were identified within the Akt gene.
• Heterozygous GA genotype at Akt2 rs4558508 was significantly associated with a higher risk of chemoresistance.

Conclusions:

• The RF model effectively predicts platinum-based NACT response in LACC.
• Key polymorphic loci in Akt2 (rs4558508, rs7259541) and Akt1 (rs1130233) are crucial for predicting NACT inefficiency.
• Akt gene variations, particularly Akt2 rs4558508, are significant predictors of chemoresistance, supporting individualized LACC therapy.