A GRAPH-BASED REPRESENTATION LEARNING APPROACH FOR BREAST CANCER RISK PREDICTION USING GENOTYPE DATA

Naga Raju Gudhe, veli-matti Kosma, Hamid Behravan, Arto Mannermaa

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Graph representation, Deep learning, Single nucleotide polymorphism, Breast cancer

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract:

Breast cancer risk prediction using genotype data is a critical task in personalized medicine. However, the high dimensionality and potential redundancy of genetic features pose challenges for accurate risk prediction. We present a graph-based representation learning pipeline for breast cancer risk prediction. Our method addresses the issue of feature redundancy by developing an ensemble-based feature selection approach. We evaluated the performance of the graph-based approach in a breast cancer risk prediction task using a dataset of 644,585 genetic variants from Biobank of Eastern Finland, consisting of 168 cases and 1558 controls and compared it with the classical machine learning models. Using 200 top-ranked genetic variants selected by the ensemble approach, the graph convolutional network
(GCN) achieved area under the ROC curve (AUC) of 0.986 ± 0.001 in discriminating cases and controls, which is better than an XGBoost model with AUC of 0.955 ± 0.0034

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6415