MetaCancer: A deep learning-based pan-cancer metastasis prediction model developed using multi-omics data

[Display omitted] Predicting metastasis in the early stages means that clinicians have more time to adjust a treatment regimen to target the primary and metastasized cancer. In this regard, several computational approaches are being developed to identify metastasis early. However, most of the approa...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Computational and structural biotechnology journal Ročník 19; s. 4404 - 4411
Hlavní autoři: Albaradei, Somayah, Napolitano, Francesco, Thafar, Maha A., Gojobori, Takashi, Essack, Magbubah, Gao, Xin
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier B.V 01.01.2021
Research Network of Computational and Structural Biotechnology
Elsevier
Témata:
Autoencoder
biotechnology
Cancer
Clinical decision support
Deep learning
DNA methylation
genome
genomics
Machine learning
Metastasis
microRNA
Multi-omics
multiomics
Pan-cancer
sequence analysis
Deep learning
Pan-cancer
Autoencoder
Machine learning
Metastasis
Clinical decision support
Multi-omics
Cancer
ISSN:2001-0370, 2001-0370
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:[Display omitted] Predicting metastasis in the early stages means that clinicians have more time to adjust a treatment regimen to target the primary and metastasized cancer. In this regard, several computational approaches are being developed to identify metastasis early. However, most of the approaches focus on changes on one genomic level only, and they are not being developed from a pan-cancer perspective. Thus, we here present a deep learning (DL)–based model, MetaCancer, that differentiates pan-cancer metastasis status based on three heterogeneous data layers. In particular, we built the DL-based model using 400 patients’ data that includes RNA sequencing (RNA-Seq), microRNA sequencing (microRNA-Seq), and DNA methylation data from The Cancer Genome Atlas (TCGA). We quantitatively assess the proposed convolutional variational autoencoder (CVAE) and alternative feature extraction methods. We further show that integrating mRNA, microRNA, and DNA methylation data as features improves our model's performance compared to when we used mRNA data only. In addition, we show that the mRNA-related features make a more significant contribution when attempting to distinguish the primary tumors from metastatic ones computationally. Lastly, we show that our DL model significantly outperformed a machine learning (ML) ensemble method based on various metrics.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:2001-0370
2001-0370
DOI:10.1016/j.csbj.2021.08.006