Abstract
Epithelial–mesenchymal transition (EMT) profoundly impacts prognosis and immunotherapy of clear cell renal cell carcinoma (ccRCC). However, not every patient is tested for EMT status because this requires additional genetic studies. In this study, we developed an EMT gene signature to classify the H&E-stained slides from The Cancer Genome Atlas (TCGA) into epithelial and mesenchymal subtypes, then we trained a deep convolutional neural network to classify ccRCC which according to our EMT subtypes accurately and automatically and to further predict genomic data and prognosis. The clinical significance and multiomics analysis of the EMT signature was investigated. Patient cohorts from TCGA (n = 252) and whole slide images were used for training, testing, and validation using an algorithm to predict the EMT subtype. Our approach can robustly distinguish features predictive of the EMT subtype in H&E slides. Visualization techniques also detected EMT-associated histopathological features. Moreover, EMT subtypes were characterized by distinctive genomes, metabolic states, and immune components. Deep learning convolutional neural networks could be an extremely useful tool for predicting the EMT molecular classification of ccRCC tissue. The underlying multiomics information can be crucial in applying the appropriate and tailored targeted therapy to the patient.
Highlights
Clear cell renal cell carcinomas account for approximately 80% of all renal cancer cases, with approximately 3.8% of all cancers in United States [1]
The RNA-Seq upper quartile normalized RSEM data was available for 539 Clear cell renal cell carcinomas (ccRCC); all data is accessible via the NCI genome data commons and the Gene Expression Omnibus
Univariate and multivariate analyses were performed and Epithelial–mesenchymal transition (EMT) gene signature was significantly associated with outcome in the multivariate analysis (Table 1)
Summary
Clear cell renal cell carcinomas (ccRCC) account for approximately 80% of all renal cancer cases, with approximately 3.8% of all cancers in United States [1]. EMT molecular stratification can predict whether patients respond to immunotherapy in several tumor types [4,5,6]. We sought to develop an EMT gene signature that can predict genomic data and prognosis of patients with ccRCC. Zhang et al presented a comprehensive morphological analysis using computer vision methods including random decision forests and artificial neural networks to establish the correlation between cellular morphological features and EMT [9]. Kather et al predicted microsatellite instability (MSI) directly from histology in gastrointestinal cancer using convolutional neural networks [10]. Unlike the typical MSI tumors, there are no standard histological criteria for EMT molecular subtypes in ccRCC patients. We investigated the deep learning neural network to precisely recognize the ccRCC EMT subtypes from whole-slide images of hematoxylin and eosin (H&E)–stained tissue from TCGA (The Cancer Genome Atlas). We compared subtype comprehensive genomic, phenotypic, and clinical data
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
