Hybrid Approach of Relation Network and Localized Graph Convolutional Filtering for Breast Cancer Subtype Classification (1711.05859v3)

Abstract: Network biology has been successfully used to help reveal complex mechanisms of disease, especially cancer. On the other hand, network biology requires in-depth knowledge to construct disease-specific networks, but our current knowledge is very limited even with the recent advances in human cancer biology. Deep learning has shown a great potential to address the difficult situation like this. However, deep learning technologies conventionally use grid-like structured data, thus application of deep learning technologies to the classification of human disease subtypes is yet to be explored. Recently, graph based deep learning techniques have emerged, which becomes an opportunity to leverage analyses in network biology. In this paper, we proposed a hybrid model, which integrates two key components 1) graph convolution neural network (graph CNN) and 2) relation network (RN). We utilize graph CNN as a component to learn expression patterns of cooperative gene community, and RN as a component to learn associations between learned patterns. The proposed model is applied to the PAM50 breast cancer subtype classification task, the standard breast cancer subtype classification of clinical utility. In experiments of both subtype classification and patient survival analysis, our proposed method achieved significantly better performances than existing methods. We believe that this work is an important starting point to realize the upcoming personalized medicine.

• The paper presents a novel hybrid deep learning model combining Relation Network and Graph CNN for improved breast cancer subtype classification.
• Experimental results using TCGA data show the hybrid model outperforms traditional methods in classification accuracy and F1 scores.
• The method demonstrates clinical relevance by distinguishing patient groups with different survival outcomes, aiding prognosis and personalized medicine.

Hybrid Approach of Relation Network and Localized Graph Convolutional Filtering for Breast Cancer Subtype Classification

Overview

The paper presents an innovative hybrid model combining Relation Network (RN) and Graph Convolutional Neural Network (graph CNN) for the classification of breast cancer subtypes. The objective is to overcome challenges in characterizing complex biological mechanisms in cancer by exploiting graph-based deep learning techniques to handle non-grid structured data. The proposed method aims to integrate relational and cooperative properties of genes to improve the classification accuracy of breast cancer subtypes.

Methodology

The authors introduce a two-stage approach to tackle breast cancer subtype classification:

1. Graph Convolutional Filtering: The graph CNN captures localized patterns within gene expression profiles, utilizing the spectral properties of graphs through Laplacian matrices. By employing Chebyshev polynomial expansion for convolution, the model efficiently processes the graph structure representing the associative gene network. The pooling strategy allows the extraction of relevant subgraphs from the adjacency matrix, contributing to the identification of cancer subtypes.
2. Relation Network: The RN is adapted to analyze the relations between gene clusters, focusing on higher-weight edge connections to infer meaningful interactions. This stage involves a custom architecture differing from traditional RN batches, using multiple layers for enhanced relational reasoning.

The integration of these components provides a robust framework for subtype classification while preserving biological insights.

Results

Experimental validation involves synthetic data, designed to simulate varying graph structures for two classes, and a real dataset from TCGA breast cancer samples using PAM50 molecular subtypes. Performance metrics, including peak and final classification accuracy and F1 scores, demonstrate the superiority of the hybrid model over traditional machine learning approaches such as SVM, Random Forest, and baseline methods (e.g., naive Bayesian classifiers).

Additionally, the visualization of feature maps through t-SNE plots reflects the clinical relevance of subtype classifications, aligning with known prognostic outcomes. Survival analysis via Kaplan-Meier plots further confirms that the method successfully distinguishes patient groups with divergent survival patterns, underscoring its practical applicability in clinical prognostics.

Implications and Future Work

The implications of this work extend to both theoretical and practical domains. Theoretically, it provides a novel integration model for handling graph-structured biological data, paving the way for further exploration in network biology and deep learning applications. Practically, the research offers promising advancements in personalized medicine, aiding in the formulation of treatment strategies based on molecular subtype profiling.

Future developments could enhance the model's capacity by incorporating multi-source data inputs such as DNA methylation and sRNA sequencing, leveraging transfer learning and multi-view learning techniques. Such enhancements could significantly contribute to more comprehensive characterization and treatment of breast cancer, as well as other complex diseases.