Back To CompGenome
Visitor Count: 70

Comprehensive Imaging-Genomics Study Of Head-Neck Squamous Cell Carcinoma


Approach

We integrated and analyzed multi-omics molecular data from The Cancer Genome Atlas (TCGA) with Computed Tomography (CT) data from The Cancer Imaging Archive (TCIA) for 126 head-neck squamous cell carcinomas.
  • Through linear regression analysis and gene set enrichment analysis, statistically significant associations (adjust p-value <= 0.05) have been identified between quantitative imaging phenotypes of tumors (such as tumor size, shape, and texture) and their corresponding molecular profiles (including miRNA expression, protein expression, somatic mutation, and transcriptional activity, copy number variation, and promoter region DNA methylation change of genetic pathway).
  • Based on quantitative imaging phenotypes, random forest classifier was used to predict the Human Papilloma Virus (HPV) status and disruptive TP53 mutation of tumor, two important factors in HNSCC diagnosis and treatment.

Results

The following figure is an overview of all statistically significant associations identified in our analysis. Each node is a genomic feature or a quantitative imaging phenotype. Each line is an identified association. Genomic features are organized into circles by data platform and indicated by different node colors. Quantitative imaging phenotypes are divided into five categories also indicated by different node colors. The node size is proportional to its connectivity relatively to other nodes in the category.
Based on quantitative imaging phenotypes, we predict HPV status and disruptive TP53 mutation of HNSCC in a five-fold cross-validation setting, and achieve average AUCs (Area Under the receiver operating characteristic Curves) of 0.71 and 0.641, respectively. This result shows the potential of using CT imaging as a non-invasive probe to detect the genomic/molecular status of HNSCC.

Download

We provide our findings as a public resource to facilitate future research on HNSCC imaging-genomics. All the identified statistically significant associations have been compiled into an excel file. [Download]

Collaborators

Dr. Clifton David Fuller's group (http://faculty.mdanderson.org/Clifton_Fuller/) at The University of Texas MD Anderson Cancer Center.
© 2019 Yuan Ji's Lab. All Rights Reserved    |    Designed by Shengjie Yang