Original Article
Unsupervised clustering reveals new prostate cancer subtypes
Abstract
Background: Prostate cancer is the second most common cancer in men. It is urgent to develop a genetic classification for prostate cancer. We aimed to establish the basis of genetic typing.
Methods: We used four series of prostate cancer data. The Cancer Genome Atlas (TCGA) RNA-Seq data were used to train the classifier. Three subgroups based on the classifier were tested whether to have significant differences in the clinical data. The other three sets were classified by the classifier and validated with respective clinical data.
Results: The classifier had 183 genes. Prostate cancer subtype 1 (PCS1) was characterized by high expression of GSTP1, with lower Gleason scores (P<0.001). PCS2 had higher Gleason score, more lymph node invasion (P=0.005) and higher pathology T stage (pT stage) (P<0.001). Three GEO (Gene Expression Omnibus) validation datasets had similar results. We even observed significances in the recurrence time among different subgroups (P=0.005 in GSE70768).
Conclusions: We established a PCS classifier (183 genes) based on RNA-Seq data, and identified three PCSs. The classification was robustly relating to clinical data which may have potential for clinical use.
Methods: We used four series of prostate cancer data. The Cancer Genome Atlas (TCGA) RNA-Seq data were used to train the classifier. Three subgroups based on the classifier were tested whether to have significant differences in the clinical data. The other three sets were classified by the classifier and validated with respective clinical data.
Results: The classifier had 183 genes. Prostate cancer subtype 1 (PCS1) was characterized by high expression of GSTP1, with lower Gleason scores (P<0.001). PCS2 had higher Gleason score, more lymph node invasion (P=0.005) and higher pathology T stage (pT stage) (P<0.001). Three GEO (Gene Expression Omnibus) validation datasets had similar results. We even observed significances in the recurrence time among different subgroups (P=0.005 in GSE70768).
Conclusions: We established a PCS classifier (183 genes) based on RNA-Seq data, and identified three PCSs. The classification was robustly relating to clinical data which may have potential for clinical use.
