Table 1.
Clinicopathological features of colon and rectal cancer patients in TCGA dataset.
Fig 1.
Construction of predictive model for CRC prognosis.
(A) Univariate cox regression analysis of the 15 genes that were related to the overall survival (OS) of patients with CRC. HR: hazard ratio. (B) Determination of the lambda value in LASSO regression model. The lambda value which corresponded to the minimum value of Partial Likelihood Deviance was considered as the optimal one. (C) Survival curves of CRC patients in TCGA dataset. P value that calculated using log-rank test was provided. (D) Time-dependent ROC curve for the patients in TCGA dataset. TP: True Positive; FP: False Positive. (E) Survival curve of CRC patients in GEO dataset. (F) Time-dependent ROC curve for the patients in GEO dataset. TP: True Positive; FP: False Positive. (G-H) Heatmaps showing the differential expressions of CX3CL1, ULK3, CDKN2A, NRG1, ATG4B, GAA, RGS19, DDIT3, GRID1, DAPK1 and SERPINA1 of the samples in two groups stratified by risk score from TCGA and GEO respectively. Horizontal and vertical axis represents samples and genes, respectively. Red and blue dot represents higher and lower expressions respectively.
Fig 2.
Risk score is a signature which could predict CRC prognosis independently.
(A) Multivariate Cox regression analysis indicated samples with HR greater than 1 was related to a higher death risk, while samples with HR less than 1 exhibited lower risk of death. (B) Survival curves of CRC patients of Stage I+StageII stratified by risk score. (C) Survival curves of CRC patients of StageIII+StageIV stratified by risk score. (D) Survival curves of CRC patients less than 67 years old stratified by risk score. (E) Survival curves of CRC patients more than 67 years old stratified by risk score. (F) Survival curves of male CRC patients stratified by risk score. (G) Survival curves of female CRC patients stratified by risk score.
Fig 3.
Nomogram for CRC survival outcome estimation.
(A) Nomogram that combined TNM stage and risk score. A single point is assigned to the stage or risk score that is perpendicular to the points line. Total point is assigned to every CRC sample by combining the sample’s risk score and TNM stage corresponded point. The OS at 1, 3 and 5 years was estimated according to the corresponding total point. (B-D) Calibration curve for the nomogram which estimated the OS at 1, 3, and 5 years respectively. (E) Time-dependent ROC curve of the samples. The horizontal axis was the false positive rate, and the vertical axis was the true positive rate. The predictive accuracy was evaluated by AUC.
Fig 4.
Immune infiltration of CRC patients with different risk scores.
(A) Landscape of relative proportions of 22 TIICs across all of the CRC samples. (B) Correlation analysis of 22 TIICs. Red: positive correlation; Blue: negative correlation; Great correlation was represented by dark color. (C) Relative proportions of 9 TIICs exhibiting significantly different infiltrating degree between CRC samples with different risk scores. (D) PCA plot of CRC samples based on the relative proportions of the 9 significantly different TIICs.
Fig 5.
Association of risk score with several important immune checkpoints as well as tumor purity.
(A) Chord diagram illustrating the correlations between mRNA levels of 6 key immune checkpoint receptors and the risk scores. The more lines between risk score and mRNAs, the greater the correlation. (B) Violin plots showing mRNA levels of the immune checkpoints which had obviously different expressions in low- and high-risk groups. (C) Violin plots showing the tumor purity in low- and high-risk groups. P value was obtained using Wilcoxon rank sum test. (D) The proportion of MSI-L + MSI-H in the high-risk group was higher than that in the low-risk group.