# Artificial neural network-based immune biomarker signature predicts pathological complete response to neoadjuvant chemotherapy in HER2-negative breast cancer

## TL;DR

A new AI model using immune-related genes can predict which HER2-negative breast cancer patients will respond well to chemotherapy before treatment.

## Contribution

Development of an artificial neural network model using five immune-related genes to predict pathological complete response to chemotherapy in HER2-negative breast cancer.

## Key findings

- The ANN model achieved an AUC of 0.858 in training and 0.773 in validation for predicting pCR.
- The five-gene signature correlates with immune cell infiltration in the tumor microenvironment.
- qRT-PCR confirmed the gene signature's association with pCR in pretreatment tumor tissues.

## Abstract

Neoadjuvant chemotherapy (NAC) is widely used in early-stage and locally advanced HER2-negative breast cancer, yet pathological complete response (pCR) is achieved in only a subset of patients. Reliable pretreatment biomarkers for predicting pCR are lacking, particularly for patients treated with standard anthracycline- and taxane-based regimens. Increasing evidence indicates that chemotherapy efficacy is closely linked to the tumor immune microenvironment, suggesting that immune-related molecular signatures may improve response prediction.

A total of 2,385 pretreatment HER2-negative breast cancer patients from ten GEO cohorts were included. GSE194040 (n = 743) was used for training, and nine independent cohorts (n = 1,642) were used for external validation. Differential expression analysis was performed separately in hormone receptor positive and negative subgroups, and genes showing concordant regulation between pCR and non-pCR cases were identified. Weighted gene co-expression network analysis (WGCNA) was applied to detect pCR-associated gene modules. Immune-related genes curated from the ImmPort database were intersected with candidate genes, followed by feature selection using least absolute shrinkage and selection operator regression, random forest, and support vector machine recursive feature elimination. An artificial neural network (ANN) model was constructed based on overlapping features and evaluated using receiver operating characteristic analysis. Immune infiltration was estimated by CIBERSORT, and transcription factor, competing endogenous RNA, and drug enrichment analyses were performed. Key genes were further validated by quantitative real-time PCR in pretreatment tumor tissues.

Five immune-related genes (CCL2, CXCL10, CXCL13, HLA-E, and IGKV1D-8) were identified as robust predictors of pCR and used to build the ANN model. The model achieved an area under the curve of 0.858(95% CI: 0.829–0.888) in the training cohort and 0.773 (95% CI: 0.735–0.808) in the external validation cohorts, demonstrating s predictive performance across independent datasets. High expression of the five-gene signature was associated with increased infiltration of cytotoxic and antigen-presenting immune cells, consistent with an immune-activated tumor microenvironment, and was confirmed by qRT-PCR analysis.

This study establishes a rigorously validated ANN-based immune gene signature for predicting response to neoadjuvant chemotherapy in HER2-negative breast cancer, providing a potential tool for pretreatment risk stratifictableation and individualized therapeutic decision-making.