Abstract
Abstract Objective: Identifying robust prognostic biomarkers of colorectal cancer (CRC) will significantly improve patients’ outcomes. However, due to the difference in detection methods, sequencing platforms and cohort populations, most of the previous molecular biomarkers were unable to cope with batch effects and rely on normalization algorithms, which might result in “test set bias”. This study aims to develop and validate a platform-independent single-sample prognostic model for colorectal cancer without the need for normalization. Methods: A total of 2510 colorectal cancer patients from 13 microarray or sequencing cohorts with complete follow-up data were included in this study. The resampling algorithm and gene pair method were combined to improve the stability of the model. After 1000 times of resampling in the training cohort, an ensemble model was constructed by selecting the k-TSP (k-Top Scoring Pairs) classifiers with high balanced accuracy. The Decision-making process in each level was generated by voting, without the need for any cutoff value. C index and D index were used to evaluate the performance of the model. Results: We constructed an ensemble model called “PIANOS”, which could significantly predict the prognosis of CRC in all 13 cohorts. The meta-estimated C index and D index of the PIANOS model were 0.74 (95%CI: 0.71-0.78, P < 0.001) and 2.77 (95%CI: 2.37-3.18, P < 0.001), respectively. After adjusting for clinicopathological factors, PIANOS was still independently associated with the prognosis of CRC. PIANOS also showed superior prognostic value when compared with previously published gene signatures or consensus molecular subtypes (CMS). GSEA analysis found that the high-risk group was enriched in angiogenesis, TGF-β and epithelial-mesenchymal transition (EMT) pathway, which was consistent with the characteristics of the CMS4 subtype in colorectal cancer. Conclusion: The PIANOS could robustly predict the prognosis of colorectal cancer across different cohorts and platforms, without the need for any normalization, which is promising for refining treatment decisions in clinical practice. Citation Format: Du Cai, Haoning Qi, Wei Wang, Min-Er Zhong, Dejun Fan, Cheng-Hang Li, Xiaojian Wu, Feng Gao. PIANOS: A platform independent and normalization free single-sample classifier for colorectal cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5126.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have