Title : Construction and validation of an unsupervised machine learning model based on RNA alternative splicing for the prognosis of non-metastatic gastric cancer
Abstract:
Background:
Alternative splicing (AS) is an important transcriptional regulatory mechanism, leading to polymorphisms in transcript and protein structure and function. This study aimed to establish more accurate prediction models for nonmetastatic gastric cancer by AS.
Methods:
The clinical and genomic information of 346 nonmetastatic gastric cancer patients in the Cancer Genome Atlas dataset (training cohort) and 25 patients in Gene Expression Omnibus dataset (validation cohort) were collected. For survival analysis, we used Cox regression to analyze the contribution of each factor to prognosis and to calculate hazard ratio and 95% confidence interval.
Results:
After dimensionality reduction of AS for all patients using the t-SNE algorithm, we used hierarchical clustering to classify all patients into two clusters. The difference in overall survival (OS) and disease-free survival (DFS) between cluster 1 and cluster 2 was statistically significant. Additionally, the differences of immune landscape between two clusters were analyzed. Further, we included cluster, age, sex, grade and stage to create multivariate nomograms for predicting OS and DFS. The precision of nomograms was evaluated by receiver operating characteristic curves and calibration curves.
Conclusion:
AS can be used as a prognostic marker in nonmetastatic gastric cancer. We established AS-based prognostic nomograms with high accuracy.
Keywords: nonmetastatic gastric cancer; machine learning; alternative splicing; disease-free survival; prognosis
Audience Take Away:
The prognostic value of alternative splicing in nonmetastatic gastric cancer patients was explored in the present study. Alternative splicing was demonstrated to a useful prognostic marker in nonmetastatic gastric cancer and prognostic nomograms integrated with alternative splicing was constructed and validated.
