Predicting pathologic complete response to neoadjuvant chemotherapy in breast cancer using sparse logistic regression

Area of Science:

• Biomedical informatics
• Genomics
• Cancer research

Background:

• Accurate prediction of treatment response is crucial in oncology.
• Previous predictive models may lack sufficient interpretability or accuracy.
• Identifying key molecular signatures can guide therapeutic strategies.

Purpose of the Study:

• To develop and evaluate two novel sparse and interpretable predictors using Sparse Logistic Regression (SLR).
• To identify informative gene signatures associated with treatment outcomes.
• To compare the predictive performance of the new models against existing methods.

Main Methods:

• Utilized Sparse Logistic Regression (SLR) to build predictive models.
• Developed SLR-65 based on 65 differentially expressed probe sets (59 genes) between Pathologic Complete Response (PCR) and Residual Disease (RD).
• Developed SLR-Notch based on 113 genes involved in Notch signaling pathways.

Main Results:

• Both SLR-65 and SLR-Notch demonstrated superior predictive performance compared to a previous study's predictor.
• SLR-65 identified 16 informative genes.
• SLR-Notch identified 12 informative genes.

Conclusions:

• Sparse Logistic Regression is effective for building interpretable and accurate predictive models in oncology.
• The identified gene signatures (SLR-65 and SLR-Notch) hold potential for predicting treatment response.
• These findings contribute to the development of personalized medicine approaches in cancer treatment.