Explaining decisions of graph convolutional neural networks: patient-specific molecular subnetworks responsible for metastasis prediction in breast cancer

Abstract

Background Contemporary deep learning approaches show cutting-edge performance in a variety of complex prediction tasks. Nonetheless, the application of deep learning in healthcare remains limited since deep learning methods are often considered as non-interpretable black-box models. However, the machine learning community made recent elaborations on interpretability methods explaining data point-specific decisions of deep learning techniques. We believe that such explanations can assist the need in personalized precision medicine decisions via explaining patient-specific predictions. Methods Layer-wise Relevance Propagation (LRP) is a technique to explain decisions of deep learning methods. It is widely used to interpret Convolutional Neural Networks (CNNs) applied on image data. Recently, CNNs started to extend towards non-Euclidean domains like graphs. Molecular networks are commonly represented as graphs detailing interactions between molecules. Gene expression data can be assigned to the vertices of these graphs. In other words, gene expression data can be structured by utilizing molecular network information as prior knowledge. Graph-CNNs can be applied to structured gene expression data, for example, to predict metastatic events in breast cancer. Therefore, there is a need for explanations showing which part of a molecular network is relevant for predicting an event, e.g., distant metastasis in cancer, for each individual patient. Results We extended the procedure of LRP to make it available for Graph-CNN and tested its applicability on a large breast cancer dataset. We present Graph Layer-wise Relevance Propagation (GLRP) as a new method to explain the decisions made by Graph-CNNs. We demonstrate a sanity check of the developed GLRP on a hand-written digits dataset and then apply the method on gene expression data. We show that GLRP provides patient-specific molecular subnetworks that largely agree with clinical knowledge and identify common as well as novel, and potentially druggable, drivers of tumor progression. Conclusions The developed method could be potentially highly useful on interpreting classification results in the context of different omics data and prior knowledge molecular networks on the individual patient level, as for example in precision medicine approaches or a molecular tumor board.

Deep learning in cancer diagnosis, prognosis and treatment selection

“…a Graph convolutional neural networks (GCNN) are designed to operate on graph-structured data. In this particular example inspired by [ 17 – 19 ], gene expression values (upper left panel) are represented as graph signals structured by a protein–protein interactions graph (lower left panel) that serve as inputs to GCNN. For a single sample (highlighted with red outline), each node represents one gene with its expression value assigned to the corresponding protein node, and inter-node connections represent known protein–protein interactions.…”

“…Cox-PASNet also correctly detected MAPK9, a gene strongly associated with GBM carcinogenesis and a novel potential therapeutic, as one the most influential genes [ 120 ]. The GCNN-explainability model from Chereda et al is the latest example of incorporating molecular networks in cancer prognosis [ 19 ]. The study used gene expression profiles, structured by a PPI from Human Protein Reference Database (HPRD) [ 121 ], to predict metastasis of breast cancer samples.…”

Graph Neural Networks and Their Current Applications in Bioinformatics

“…Take learning molecular fingerprint ( Duvenaud et al, 2015 ) as an example; the fingerprint encoding method using neural graphs can take into account the similarity between molecular fragments to achieve a more meaningful feature representation, which is also ignored in traditional fingerprint encoding. In the prediction of metastasis for breast cancer patients ( Chereda et al, 2021 ), the graph layer-wise relevance propagation was proposed to explain how GCN generates predictions based on patient-specific PPI sub-network data which could be potentially highly useful for the development of personalized medicine. In histological image analysis, Sureka et al (2020) modeled histological tissue as a nuclear graph and established a graph convolutional network framework based on attention mechanism and node occlusion for disease diagnosis.…”

“…In order to solve the over-fitting problem, Cai et al (2020) proposed a new GCN model based on fine-grained edge dropout and coarse-grained node dropout to reduce the over-fitting in sparse graphs. Chereda et al (2021) combined the PPI network and gene expression data for patients and utilized GCN to classify the nodes in the patient’s sub-network for predicting breast cancer metastasis. In the classification of breast cancer subtypes, there are also related studies based on local GCN, which was used to combine with the PPI network and the gene expression matrix information of multiple patients ( Rhee et al, 2018 ).…”

Towards Clinical Application of Artificial Intelligence in Ultrasound Imaging

“…It is necessary to assess the input data quality and to accumulate robust technologies, including effective data structuring and algorithm development, to facilitate the clinical implementation of AI devices. Another concern is the AI black box problem, whereby the decision-making process of the manner in which complicated synaptic weighting is performed in the hidden layers of CNNs is unclear [28]. Examiners need to understand and explain the rationale for diagnosis to patients objectively for obtaining informed consent in constructing valid AI-based US diagnostic technologies in clinical practice.…”