As a common malignant tumor of the digestive tract, colon cancer has the characteristics of a high incidence rate, frequent postoperative recurrence and high mortality. At present, pathophysiological evaluation, treatment decision-making and prognosis evaluation of colon cancer mainly depend on cancer cell-centered evaluation factors, such as the TNM staging system and serum molecular markers, such as CEA and CA19-9 [14, 15]. Although many gene prognostic indicators have been used to predict the prognosis of patients with colon cancer, their accuracy is still uncertain [16, 17]. Therefore, it is urgent to explore reliable prognostic biomarkers to predict the survival of COAD patients.
Interestingly, lncRNAs can regulate the expression of coding genes by affecting adjacent genes or distant genes on other chromosomes . Because of the obvious abnormal expression of lncRNAs in the tumors of patients with colon cancer, lncRNAs such as SNHG7 , SNHG11 and POU6F2  are highly specific and sensitive potential biomarkers. As a specific biomarker of colon cancer, lncRNAs have prominent effects on diagnosis, curative effect prediction and prognosis. Importantly, it was suggested that the accuracy of a cancer diagnosis model with a combination of multiple biomarkers was significantly better than that of a single biomarker prediction model . Due to the biological heterogeneity of tumors and the technical deviation caused by the use of different sequencing platforms, previous risk models of colon cancer need to be standardized to gene expression profiles. To achieve robustness of prediction, we used a new method without the need to consider the data deviation arising from the use of different platforms. Moreover, the new analysis method does not require the specific expression level of lncRNAs or the scaling and normalization of the expression levels of lncRNAs but rather uses relative ranking and paired comparisons of lncRNA expression levels. This method obtained reliable results in a previous study of human hepatocellular carcinoma .
In this study, we identified a prognostic DEirlncRNAP model to predict the prognosis of COAD patients. The prognostic model consisted of 71 DEirlncRNAPs, including 90 independent DEirlncRNAs. It has been shown that lncRNAs are key regulators of the immune response, participate in gene activation and regulate the immunophenotype [23, 24]. With the remarkable achievements of immunotherapy for cancer treatment, irlncRNAs have gradually become a new hot spot and have been found to be prognostic factors for breast cancer , renal cancer  and pancreatic cancer .
A model for predicting the prognosis of colon cancer with 10 lncRNAs was established, which suggested that lncRNAs with high expression levels have important biological functions . Wu et al.  created a prognostic model composed of 12 irlncRNAs, and the internal verification C index of the nomogram was 0.807. We obtained DEirlncRNAPs by differential coexpression analysis and established a DEirlncRNAP model by modified Lasso penalized regression. This algorithm only requires analysis of the expression level of paired lncRNAs, and it is not necessary to calculate the specific expression level of each lncRNA, so it has higher clinical practical value than other models. Some DEirlncRNAs involved in our study have been found to play an important role in the occurrence, progression and metastasis of various cancers, such as CDKN2B-AS1 in lung cancer and cervical cancer [30, 31], MNX1-AS1 in esophageal squamous cell carcinoma and gastric cancer [32, 33], B4GALT1-AS1 in colon cancer , NKILA in hepatocellular carcinoma and breast cancer [35, 36] and MIR17HG in glioma and osteosarcoma [37, 38]. Among them, the expression of NKILA in colorectal cancer was lower than that in normal tissues and adenomas . MNX1-AS1 regulates the miR-218-5p/SEC61A1 axis to promote the development of colon cancer . Both B4GALTI-AS1 and MIR17HG promote the migration of colon cancer cells [34, 41]. The abnormal expression of these lncRNAs facilitates the development of colon cancer. We are the first to identify the other lncRNAs used to build this model that may be associated with cancer and could be novel biomarkers for further colon cancer research.
We used seven common methods to evaluate the difference in infiltrating immune cells between the high- and low-risk groups, including XCELL, TIMER, QUANTISEQ, MCPCOUNTER, EPIC, CIBERSORT-ABS and CIBERSORT. Because the algorithms of each platform are complex and different, we integrated the results. The results suggested that high risk positively correlated with hematopoietic stem cells, macrophages, monocytes, endothelial cells and cancer-associated fibroblast cells and negatively correlated with class-switch memory B cells, plasmacytoid dendritic cells, CD8+ T cells and mast cells. It has been reported that endothelial cells can form new tumor blood vessels and provide nutrition for tumors . Macrophages induce the proliferation and migration of colon cancer cells . Monocytes have also been found to be closely associated with liver metastasis of colon cancer . Plasmacytoid dendritic cells produce type I interferon, which has a strong immune response to tumors and contributes to the establishment of an immunosuppressive tumor microenvironment . Notably, tumor-infiltrating CD8+ T cells often indicate a better immunotherapy response and prognosis . It was also found that the prognosis of high-risk patients with tumor-infiltrating immune cells was poor in the DEirlncRNAP model. Galon et al.  proposed that an immune score based on tumor immune cell density analysis can predict the prognosis of patients and that this method is more accurate than TNM analysis.
This immune score also contributes to individualized immunotherapy and chemotherapy in the clinic [48, 49]; thus, we analyzed the IC50 of chemotherapy drugs and immune checkpoint inhibitors in the high-risk group. The results showed that the high-risk group was more sensitive to gefitinib, thapsigargin and vinorelbine than to 5-fluorouracil and irinotecan. Liua et al  found that vinorelbine can inhibit the metastasis of colon cancer cells by reducing metastasis, invasion and epithelial-mesenchymal transition. This model may provide a new direction for the further study of chemotherapeutic drugs for colon cancer. However, we believe that immunotherapy can eliminate cancer cells and produce new antigens, which may have an advantage over chemotherapy in inhibiting tumor development. Although the results of our model and immune checkpoint inhibitor analysis only showed that the difference in the expression of CTLA4 was statistically significant, the tumor-infiltrating immune cells and immunophenotype of the patients were different, and the expression of the verified biomarkers should also be different. Due to the heterogeneity of tumors, we need to effectively and accurately identify the tumor type, tumor invasion matrix, and molecular and functional characteristics of COAD, identify more biomarkers, develop different immunotherapy regimens, and carry out personalized precision treatment.
However, there are still some deficiencies in our research. The TCGA data we adopted have some defects, mainly because the data are only from a small number of countries, the sample size is limited, the data are not updated in a timely manner, and the sequencing technologies and quality control methods are different across the samples, all of which have a certain impact on the accuracy of the data. Moreover, we only searched the RNA-seq dataset, which does not fully represent the status of the tumor genes, and we were not able to collect the lncRNA expression level data of the colon cancer patients. The same dataset was used to build and verify the model. The prognostic efficacy and potential mechanism of this model still need to be externally verified through the analysis of real clinical data and basic research. Therefore, we used a 0-or-1 matrix to screen the qualified DEirlncRNAPs without calculating the expression level of lncRNAs to minimize sample error. In addition, the algorithm was verified through several parameters, such as the survival time of the patients, clinicopathological features, tumor infiltrating immune cells and clinical therapeutics. Nevertheless, external validation with clinical data is still needed to confirm the reliability of the model. Therefore, more functional studies and in vitro and in vivo experiments should be carried out to test the accuracy of the DEirlncRNAP risk assessment model to improve its clinical applicability.