ABHISHEK BHANSALI 2019B1A41468H

"Unlocking the Complexity of Breast Cancer: Machine Learning

Model Predicts Response to Neoadjuvant Therapy"

Breast cancer is a complex disease that involves a diverse ecosystem of malignant cells and
microenvironmental factors. The composition and interactions within these ecosystems are
essential contributors to the response to treatment, and efforts to develop response predictors
have not yet incorporated this knowledge. This study aimed to collect multi-omic data from pre-
treatment biopsies of breast tumors to investigate the relationship between pre-treatment
features and response to chemotherapy with or without HER2-targeted therapy before surgery.

To predict a complete pathological response, the researchers developed a multi-omic machine

learning model that achieved an area under the curve of 0.87 in an external validation cohort.
Using a machine learning framework, the model incorporates clinical, digital pathology,
genomic, and transcriptome variables. Using various feature combinations, the authors created
six prediction models with 34 predictive features.

The study found that response to neoadjuvant therapy was connected to traits discovered
through clinical, molecular, and digital pathology investigations. The most common subtypes of
breast cancer were represented in the data set. The models' areas under the curve varied
between 0.70 and 0.87 based on the qualities they were designed to predict. The final model
included clinical, DNA/RNA, digital pathology, and therapy features, with an area under the
curve of 0.87.

The study also discovered that proliferation is a crucial determinant of response to therapy and
that the tumor environment's initial features largely govern this response. In HER2+
malignancies, the response to chemotherapy and HER2-targeted antibodies is independent of
the cancer cell's ability to proliferate. Clonal variety and subclonal mutations have been
connected to the persistence of the disease.

The tumor immune microenvironment (TiME) is a significant determinant in predicting the

response to therapy in previously untreated malignancies. Tumors with a robust immune
infiltration in the tumor microenvironment (TME) were more likely to respond favorably to
treatment than those with a dampened immune response.

The findings suggest that machine learning models incorporating clinical, molecular, and digital
pathology data can better predict therapy response than models employing clinical variables.
The study also suggests that the models could pave the way for using molecular and digital
pathology to choose medicines for future clinical trials, including adjuvant medications.
In summary, the study found that analyzing biopsies before treatment can predict a patient's
response to chemotherapy or HER2-targeted therapy. Machine learning models incorporating
clinical, molecular, and digital pathology data can better predict therapy response than models
ABHISHEK BHANSALI 2019B1A41468H

employing clinical variables. The study's findings may pave the way for using molecular and
digital pathology to choose medicines for future clinical trials, including adjuvant medications.