A transcriptome-Based Deep Neural Network Classifier for Identifying the Site of Origin in Mucinous Cancer.

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Mucinous cancers present a diagnostic challenge due to difficulties in pinpointing their site of origin.
• Existing diagnostic tools are insufficient for precise origin identification in mucinous carcinomas.

Purpose of the Study:

• To evaluate the efficacy of a novel transcriptome-based classifier for determining the primary site of mucinous cancers.
• To develop a computational tool for improving the diagnosis of mucinous cancer origins.

Main Methods:

• Utilized transcriptomic data from The Cancer Genome Atlas (TCGA) for 1878 non-mucinous and 82 mucinous cancer specimens across 7 sites.
• Developed a deep neural network classifier trained on 427 differentially expressed genes.
• Validated the classifier using independent training, validation, and a test set of 14 newly sequenced mucinous cancer specimens.

Main Results:

• The transcriptome-based classifier achieved high accuracy: 0.998 (training), 0.939 (validation), and 0.857 (test set).
• t-distributed Stochastic Neighbor Embedding (t-SNE) analysis confirmed that test set samples clustered with known origins from the training set.
• The classifier demonstrated robust performance in identifying the site of origin for mucinous cancers, even in newly analyzed samples.

Conclusions:

• A transcriptome-based classifier can accurately identify the site of origin in mucinous cancer.
• This approach offers a promising tool for diagnosing cancers of unknown primary site, despite limitations of small test set size.