A disease network‐based deep learning approach for characterizing melanoma

Abstract

Multiple types of genomic variations are present in cutaneous melanoma and some of the genomic features may have an impact on the prognosis of the disease. The access to genomics data via public repositories such as The Cancer Genome Atlas (TCGA) allows for a better understanding of melanoma at the molecular level, therefore making characterization of substantial heterogeneity in melanoma patients possible. Here, we proposed an approach that integrates genomics data, a disease network, and a deep learning model to classify melanoma patients for prognosis, assess the impact of genomic features on the classification and provide interpretation to the impactful features. We integrated genomics data into a melanoma network and applied an autoencoder model to identify subgroups in TCGA melanoma patients. The model utilizes communities identified in the network to effectively reduce the dimensionality of genomics data into a patient score profile. Based on the score profile, we identified three patient subtypes that show different survival times. Furthermore, we quantified and ranked the impact of genomic features on the patient score profile using a machine-learning technique. Follow-up analysis of the top-ranking features provided us with the biological interpretation of them at both pathway and molecular levels, such as their mutation and interactome profiles in melanoma and their involvement in pathways associated with signaling transduction, immune system and cell cycle. Taken together, we demonstrated the ability of the approach to identify disease subgroups using a deep learning model that captures the most relevant information of genomics data in the melanoma network.

Artificial intelligence in cancer target identification and drug discovery

“…The other is their ability to effectively integrate large and diverse data. It is possible for ML-based networks biology analysis algorithms to integrate multiomics biological network data and identify novel targets 263 , because of the fast development of deep learning models and the easy access to high-throughput biological.…”

“…The third challenge is hard to provide interpretability of deep learning models 185 . However, a recent study sheds a light to resolve the issue through a combination of a disease network with a neural network to characterize the mechanism of melanoma 263 . In addition, graphs-based neural networks can improve the interpretability of deep learning models 265 .…”

Computational Intelligence-Based Melanoma Detection and Classification Using Dermoscopic Images

“…Lai et al [ 7 ] presented the technique which combines genomics data, a disease network, and the DL technique for classifying melanoma patients to prognosis, evaluating the influence of genomic features on the classifier, and offering interpretation to impactful features. It combined genomics data with a melanoma network and executed the AE method for identifying subgroups from TCGA melanoma patients.…”

“…To identify skin cancer quickly at the beginning and resolve the abovementioned problems, there is a comprehensive study solution by proposing a computer image analysis algorithm [ 6 ]. Most of the algorithmic solution was parametric, which means they needed information to be distributed normally [ 7 ]. Since the nature of information could not be controlled, this method will be inadequate to precisely identify the disease.…”

Integration of lncRNAs, Protein-Coding Genes and Pathology Images for Detecting Metastatic Melanoma

“…OMIC data provide a molecular-level view of melanoma, and the detected biomarkers facilitate the understanding of the onset and progression mechanisms of melanoma. Lai et al selected the fully connected melanoma subnetwork with the best modularity score and proposed an autoencoder-based deep learning network to detect different melanoma subgroups using the genomic data in The Cancer Genome Atlas [ 26 ]. Wei revealed 798 differentially expressed genes of melanoma and built a support vector machine (SVM)-based classifier using the top 110 biomarker genes to achieve at least 0.944 in accuracy across three independent datasets [ 27 ].…”