Dataset Information

A machine learning-based prognostic predictor for stage III colon cancer.

ABSTRACT: Limited biomarkers have been identified as prognostic predictors for stage III colon cancer. To combat this shortfall, we developed a computer-aided approach which combing convolutional neural network with machine classifier to predict the prognosis of stage III colon cancer from routinely haematoxylin and eosin (H&E) stained tissue slides. We trained the model by using 101 cancers from West China Hospital (WCH). The predictive effectivity of the model was validated by using 67 cancers from WCH and 47 cancers from The Cancer Genome Atlas Colon Adenocarcinoma database. The selected model (Gradient Boosting-Colon) provided a hazard ratio (HR) for high- vs. low-risk recurrence of 8.976 (95% confidence interval (CI), 2.824-28.528; P, 0.000), and 10.273 (95% CI, 2.177-48.472; P, 0.003) in the two test groups, from the multivariate Cox proportional hazards analysis. It gave a HR value of 10.687(95% CI, 2.908-39.272; P, 0.001) and 5.033 (95% CI,1.792-14.132; P, 0.002) for the poor vs. good prognosis groups. Gradient Boosting-Colon is an independent machine prognostic predictor which allows stratification of stage III colon cancer into high- and low-risk recurrence groups, and poor and good prognosis groups directly from the H&E tissue slides. Our findings could provide crucial information to aid treatment planning during stage III colon cancer.

SUBMITTER: Jiang D

PROVIDER: S-EPMC7316723 | biostudies-literature | 2020 Jun

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

A machine learning-based prognostic predictor for stage III colon cancer.

Jiang Dan D Liao Junhua J Duan Haihan H Wu Qingbin Q Owen Gemma G Shu Chang C Chen Liangyin L He Yanjun Y Wu Ziqian Z He Du D Zhang Wenyan W Wang Ziqiang Z

Scientific reports 20200625 1

Limited biomarkers have been identified as prognostic predictors for stage III colon cancer. To combat this shortfall, we developed a computer-aided approach which combing convolutional neural network with machine classifier to predict the prognosis of stage III colon cancer from routinely haematoxylin and eosin (H&E) stained tissue slides. We trained the model by using 101 cancers from West China Hospital (WCH). The predictive effectivity of the model was validated by using 67 cancers from WCH ...[more]

PMID: 32587295

Similar Datasets

Project description:Background The identification of high-risk stage II colon cancers is key to the selection of patients who require adjuvant treatment after surgery. Microarray-based multigene-expression signatures derived from stem cells and progenitor cells hold promise, but they are difficult to use in clinical practice. Methods We used a new bioinformatics approach to search for biomarkers of colon epithelial differentiation across gene-expression arrays and then ranked candidate genes according to the availability of clinical-grade diagnostic assays. With the use of subgroup analysis involving independent and retrospective cohorts of patients with stage II or stage III colon cancer, the top candidate gene was tested for its association with disease-free survival and a benefit from adjuvant chemotherapy. Results The transcription factor CDX2 ranked first in our screening test. A group of 87 of 2115 tumor samples (4.1%) lacked CDX2 expression. In the discovery data set, which included 466 patients, the rate of 5-year disease-free survival was lower among the 32 patients (6.9%) with CDX2-negative colon cancers than among the 434 (93.1%) with CDX2-positive colon cancers (hazard ratio for disease recurrence, 3.44; 95% confidence interval [CI], 1.60 to 7.38; P=0.002). In the validation data set, which included 314 patients, the rate of 5-year disease-free survival was lower among the 38 patients (12.1%) with CDX2 protein-negative colon cancers than among the 276 (87.9%) with CDX2 protein-positive colon cancers (hazard ratio, 2.42; 95% CI, 1.36 to 4.29; P=0.003). In both these groups, these findings were independent of the patient's age, sex, and tumor stage and grade. Among patients with stage II cancer, the difference in 5-year disease-free survival was significant both in the discovery data set (49% among 15 patients with CDX2-negative tumors vs. 87% among 191 patients with CDX2-positive tumors, P=0.003) and in the validation data set (51% among 15 patients with CDX2-negative tumors vs. 80% among 106 patients with CDX2-positive tumors, P=0.004). In a pooled database of all patient cohorts, the rate of 5-year disease-free survival was higher among 23 patients with stage II CDX2-negative tumors who were treated with adjuvant chemotherapy than among 25 who were not treated with adjuvant chemotherapy (91% vs. 56%, P=0.006). Conclusions Lack of CDX2 expression identified a subgroup of patients with high-risk stage II colon cancer who appeared to benefit from adjuvant chemotherapy. (Funded by the National Comprehensive Cancer Network, the National Institutes of Health, and others.).

Project description:BackgroundColon cancer is one of the leading causes of cancer deaths in the USA and around the world. Molecular level characters, such as gene expression levels and mutations, may provide profound information for precision treatment apart from pathological indicators. Transcription factors function as critical regulators in all aspects of cell life, but transcription factors-based biomarkers for colon cancer prognosis were still rare and necessary.MethodsWe implemented an innovative process to select the transcription factors variables and evaluate the prognostic prediction power by combining the Cox PH model with the random forest algorithm. We picked five top-ranked transcription factors and built a prediction model by using Cox PH regression. Using Kaplan-Meier analysis, we validated our predictive model on four independent publicly available datasets (GSE39582, GSE17536, GSE37892, and GSE17537) from the GEO database, consisting of 925 colon cancer patients.ResultsA five-transcription-factors based predictive model for colon cancer prognosis has been developed by using TCGA colon cancer patient data. Five transcription factors identified for the predictive model is HOXC9, ZNF556, HEYL, HOXC4 and HOXC6. The prediction power of the model is validated with four GEO datasets consisting of 1584 patient samples. Kaplan-Meier curve and log-rank tests were conducted on both training and validation datasets, the difference of overall survival time between predicted low and high-risk groups can be clearly observed. Gene set enrichment analysis was performed to further investigate the difference between low and high-risk groups in the gene pathway level. The biological meaning was interpreted. Overall, our results prove our prediction model has a strong prediction power on colon cancer prognosis.ConclusionsTranscription factors can be used to construct colon cancer prognostic signatures with strong prediction power. The variable selection process used in this study has the potential to be implemented in the prognostic signature discovery of other cancer types. Our five TF-based predictive model would help with understanding the hidden relationship between colon cancer patient survival and transcription factor activities. It will also provide more insights into the precision treatment of colon cancer patients from a genomic information perspective.

Project description:Colorectal cancer (CRC) is the third leading cause of cancer-related death and the fourth most commonly diagnosed cancer worldwide. Due to a lack of diagnostic biomarkers and understanding of the underlying molecular mechanisms, CRC's mortality rate continues to grow. CRC occurrence and progression are dynamic processes. The expression levels of specific molecules vary at various stages of CRC, rendering its early detection and diagnosis challenging and the need for identifying accurate and meaningful CRC biomarkers more pressing. The advances in high-throughput sequencing technologies have been used to explore novel gene expression, targeted treatments, and colon cancer pathogenesis. Such approaches are routinely being applied and result in large datasets whose analysis is increasingly becoming dependent on machine learning (ML) algorithms that have been demonstrated to be computationally efficient platforms for the identification of variables across such high-dimensional datasets. We developed a novel ML-based experimental design to study CRC gene associations. Six different machine learning methods were employed as classifiers to identify genes that can be used as diagnostics for CRC using gene expression and clinical datasets. The accuracy, sensitivity, specificity, F1 score, and area under receiver operating characteristic (AUROC) curve were derived to explore the differentially expressed genes (DEGs) for CRC diagnosis. Gene ontology enrichment analyses of these DEGs were performed and predicted gene signatures were linked with miRNAs. We evaluated six machine learning classification methods (Adaboost, ExtraTrees, logistic regression, naïve Bayes classifier, random forest, and XGBoost) across different combinations of training and test datasets over GEO datasets. The accuracy and the AUROC of each combination of training and test data with different algorithms were used as comparison metrics. Random forest (RF) models consistently performed better than other models. In total, 34 genes were identified and used for pathway and gene set enrichment analysis. Further mapping of the 34 genes with miRNA identified interesting miRNA hubs genes. We identified 34 genes with high accuracy that can be used as a diagnostics panel for CRC.

Project description:The prognostic value of BRAF and KRAS mutations within microsatellite-unstable (MSI) and microsatellite-stable (MSS) subgroups of resected colon carcinoma patients remains controversial. We examined this question in prospectively collected biospecimens from stage III colon cancer with separate analysis of MSI and MSS tumors from patients receiving adjuvant FOLFOX +/- cetuximab in two adjuvant therapy trials.Three groups were defined: BRAF Mutant, KRAS Mutant, and double wild-type. The analytic strategy involved estimation of study-specific effects, assessment of homogeneity of results, and then analysis of pooled data as no differences in patient outcome were found between treatment arms in both trials. Associations of mutations with patient outcome were analyzed, and multivariable models were adjusted for treatment and relevant factors.Four thousand four hundred eleven tumors were evaluable for BRAF and KRAS mutations and mismatch repair status; 3934 were MSS and 477 were MSI. In MSS patients, all BRAF V600E mutations (hazard ratio [HR] = 1.54, 95% confidence interval [CI] = 1.23 to 1.92, P < .001), KRAS codon 12 alterations, and p.G13D mutations (HR?=?1.60, 95% CI?=?1.40 to 1.83, P < .001) were associated with shorter time to recurrence (TTR) and shorter survival after relapse (SAR; HR?=?3.02 , 95% CI?=?2.32 to 3.93, P < .001, and HR?=?1.20, 95% CI?=?1.01 to 1.44, P = .04, respectively). Overall survival (OS) in MSS patients was poorer for BRAF-mutant patients (HR?=?2.01, 95% CI?=?1.56 to 2.57, P < .001) and KRAS-mutant patients (HR?=?1.62, 95% CI?=?1.38 to 1.91, P < .001) vs wild-type. No prognostic role of KRAS or BRAF mutations was seen in MSI patients. Furthermore, no interaction was found between treatment arm (with or without cetuximab) and KRAS and BRAF mutations for TTR or OS in MSS patients.In a pooled analysis of resected stage III colon cancer patients receiving adjuvant FOLFOX, BRAF or KRAS mutations are independently associated with shorter TTR, SAR, and OS in patients with MSS, but not MSI, tumors. Future clinical trials in the adjuvant setting should consider these mutations as important stratification factors.

Dataset Information

A machine learning-based prognostic predictor for stage III colon cancer.

Publications

A machine learning-based prognostic predictor for stage III colon cancer.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets