A novel computational framework for integrating multidimensional data to enhance accuracy in predicting the prognosis of colorectal cancer. Issue 2 (18th December 2022)

• Record Type: Journal Article
• Authors: Zhang, Qinran; Xu, Yuhong; Kang, Shiyang; Chen, Junquan; Yao, Zhihao; Wang, Haitao; Wu, Qinian; Zhao, Qi; Zhang, Qihua; Xu, Rui‐hua; Zou, Xiufen; Luo, Huiyan
• Is Part Of: MedComm. Volume 1:Issue 2(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-12-18
• Subjects: biomarker -- multidimensional data integration -- prognosis prediction -- stage II/III colorectal cancer -- the random non‐negative matrix factorization algorithm; Medical innovations -- Periodicals; Medicine -- Research -- Periodicals; Biology -- Research; Medicine -- Research; Periodicals; 610.72
• DOI: 10.1002/mef2.27
• Languages: English
• ISSNs: 2769-6456

Abstract

Abstract: Accurate prognosis prediction is the key to achieving precision treatment and guiding the selection of adjuvant chemotherapy in high‐risk stage II/III colorectal cancer (CRC) patients. Here we developed a novel machine learning method, the random non‐negative matrix factorization (RNMF) algorithm, which outperformed traditional non‐negative matrix factorization in terms of computational speed, accuracy, and robustness in simulated data sets. Moreover, based on multidimensional data from CRC patients from The Cancer Genome Atlas database and DNA methylation data from those from Sun Yat‐sen University cancer center, we further demonstrated the excellent performance of a novel prognostic computational framework based on the RNMF (PCF_RNMF), which is capable of integrating multidimensional training while allowing survival prediction when single dimensional data for validation is provided. This novel algorithm has great potential to mitigate the challenge of integrating various types of data in public databases with clinically available single‐dimensional data to allow cost‐effective survival prediction for CRC patients in clinical practice. Abstract : The computation framework prognostic computational framework is based on random non‐negative matrix factorization (RNMF). (a) The multidimensional data in TCGA were used to select features. (b) The RNMF method was used to obtain the survival score matrix and the prediction coefficient matrix. (c) The prognosis model was built by using Cox stepwise regression. (d) The corresponding prognostic score can be obtained based on the prediction coefficient matrix of the methylation dimension by RNMF in (b).