Multi-modal characteristics of LncRNA-derived subtypes in colorectal cancer

The growing knowledge of long non-coding RNA (LncRNA) has suggested the role and predictive potential of LncRNA in cancer, but has not been translated into effective practical tools. The morbidity of colorectal cancer (CRC) has been declining because of effective screening tools. However, the mechanism of the disease is not well understood. The LncRNAs that extracted using univariable Cox analysis were explored for the potential functions and applied for classification of LncRNA-derived subtypes of CRC. Multi-modal data involving microscope to macroscope levels were applied to evaluate the molecular, cell, and medical image characteristics of LncRNA-derived subtypes. Radiomic signature calculated by LASSO regression analysis was utilized for distinguishing the CRC subtypes. The performance of LncRNA and radiomic signature were assessed with respect to its calibration, discrimination, and clinical usefulness. The LncRNA signature composed of 8 LncRNAs significantly associated with prognosis, metabolism, and organismal system showed good performance in patients’ prognosis assessment (AUC, 0.717; time-dependent ROC > 0.71). Patients separated into two subtypes (HRG vs. LRG) based on the LncRNA signature displayed notable prognostic differences (Kaplan-Meier, p < 0.001). Multi-modal analysis indicated obvious heterogeneity between HRG and LRG subgroups, mainly focused on the abnormal enrichment of glycerophospholipid metabolism pathway, different cellular differentiation, and density level characteristics in the tumor region. In addition, the radiomic signature showed good accuracy (AUC, 0.919), discrimination (p < 0.05), and generalization ability (external independent test cohort, p < 0.05) for LncRNA-derived subgroups assessment. This study described the multi-modal characteristics landscape of LncRNA-derived subtypes in CRC and strengthened the connection from microscope to macroscope levels.