Brain Cancer Prediction Using Machine Learning Methods and High-Throughput Molecular Data

Knowledge of 1p/19q-codeletion and IDH1/2 mutational status is necessary to interpret any investigational study of diffuse gliomas in the modern era. While DNA sequencing is the gold standard for determining IDH mutational status, genome-wide methylation arrays and gene expression profiling have been used for surrogate mutational determination. Previous studies by our group suggest that 1p/19q-codeletion and IDH mutational status can be predicted by genome-wide somatic copy number alteration (SCNA) data alone, however a rigorous model to accomplish this task has yet to be established. In this study, we used SCNA data from 786 adult diffuse gliomas in The Cancer Genome Atlas (TCGA) to develop a two-stage classification system that identifies 1p/19q-codeleted oligodendrogliomas and predicts the IDH mutational status of astrocytic tumors using a machine-learning model. Cross-validated results on TCGA SCNA data showed near perfect classification results. Furthermore, our astrocytic IDH mutation model validated well on four additional datasets (AUC = 0.97, AUC = 0.99, AUC = 0.95, AUC = 0.96) as did our 1p/19q-codeleted oligodendroglioma screen on the two datasets that contained oligodendrogliomas (MCC = 0.97, MCC = 0.97). We then retrained our system using data from these validation sets and applied our system to a cohort of REMBRANDT study subjects for whom SCNA data, but not IDH mutational status, is available. Overall, using genome-wide SCNAs, we successfully developed a system to robustly predict 1p/19q-codeletion and IDH mutational status in diffuse gliomas. This system can assign molecular subtype labels to tumor samples of retrospective diffuse glioma cohorts that lack 1p/19q-codeletion and IDH mutational status, such as the REMBRANDT study, recasting these datasets as validation cohorts for diffuse glioma research.

Background: Characterization of focal somatic copy number alterations (SCNAs) has led to the identification of many cancer genes, yet similar investigations of arm-level SCNAs remain challenging. The identity of driver genes within these broad SCNAs remains a critical unanswered question in cancer genetics. One of the most frequent arm-level SCNAs is 9p loss, which contains the tumor suppressor gene (TSG) CDKN2A. It is also believed that other TSGs exist on 9p, though their identity has yet to be revealed. Methods: We analyzed every arm-level SCNA in 28 cancer types from The Cancer Genome Atlas (TCGA) to clarify the relative impact of 9p loss across cancer. We also performed a multi-tiered genomic dissection of chromosome 9p in 540 patients from 3 independent lower grade glioma (LGG) datasets (TCGA, REMBRANDT, MSKCC) to pinpoint genetic loci that are tied to tumor aggressiveness and poor survival. Focal and arm-level SCNAs were determined by GISTIC2.0. As per recently proposed criteria, 3 LGG subtypes were clustered by IDH mutation and 1p/19q deletion status. Survival analyses were performed using log-rank tests or Cox regression. Results: We found that chromosome 9p loss is one of the most frequent and prognostic arm-level events across 28 TCGA cancer types. Of these, the strongest 9p survival association was found in LGG. We performed a large-scale associations test of 376 important clinical and molecular variables and revealed that 9p loss was only associated with 6q or 9q loss, and 9p loss alone was sufficient to predict poor prognosis. On subtype-specific analysis, 9p loss predicted worse overall survival (OS) in both IDH mutant LGG subtypes but not IDH wild-type (IDHwt). To identify underlying drivers on 9p, we identified 87 genes most frequently targeted by 9p loss. Additional genetic events were rare except for homozygous deletion (HD) at 9p21.3, which contains CDKN2A. In contrast, mRNA/miRNA expression at all gene loci was much more variable and poorly correlated to copy number status. Therefore, pan-LGG and subtype-specific gene expression analyses were used to identify several drivers of tumor aggressiveness, notably KLHL9, MTAP, PLAA, and PTPRD. Although CDKN2A HD was linked to worse OS in a pan-LGG analysis, this event significantly co-occurred with the LGG IDHwt subtype, a group with GBM-like survival outcomes. Further, CDKN2A copy number status and expression did not predict worse OS in any subtype. Discussion: We characterize the nature of 9p loss in LGG, pinpoint genes involved in worse survival, and redefine the role of the tumor suppressor CDKN2A. These results provide new critical insight into long-standing questions about the nature of chromosome 9p loss and establish a framework for examining other arm-level SCNAs in cancer. Citation Format: David M. Roy, Logan A. Walsh, Alexis Desrichard, Jianjiong Gao, Promita Bose, Jason T. Huse, William Lee, Timothy A. Chan. Multidimensional genomic dissection of chromosome 9p in glioma. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr LB-169. doi:10.1158/1538-7445.AM2015-LB-169

Development and validation of an artificial intelligence system for triple-negative breast cancer identification and prognosis prediction: a multicentre retrospective study

Genomic features, including tumor mutation burden (TMB), microsatellite instability (MSI), and somatic copy number alteration (SCNA), had been demonstrated to be involved with the tumor microenvironment (TME) and outcome of gastric cancer (GC). We obtained profiles of TMB, MSI, and SCNA by processing 405 GC data from The Cancer Genome Atlas (TCGA) and then conducted a comprehensive analysis though “iClusterPlus.” A total of two subgroups were generated, with distinguished prognosis, somatic mutation burden, copy number changes, and immune landscape. We revealed that Cluster1 was marked by a better prognosis, accompanied by higher TMB, MSIsensor score, TMEscore, and lower SCNA burden. Based on these clusters, we screened 196 differentially expressed genes (DEGs), which were subsequently projected into univariate Cox survival analysis. We constructed a 9-gene immune risk score (IRS) model using LASSO-penalized logistic regression. Moreover, the prognostic prediction of IRS was verified by receiver operating characteristic (ROC) curve analysis and nomogram plot. Another independent Gene Expression Omnibus (GEO) contained specimens from 109 GC patients was designed as an external validation. Our works suggested that the 9‐gene‐signature prediction model, which was derived from TMB, MSI, and SCNA, was a promising predictive tool for clinical outcomes in GC patients. This novel methodology may help clinicians uncover the underlying mechanisms and guide future treatment strategies.

Glioblastoma multiforme (GBM) is the most devastating brain tumor with less than 5% of patients surviving 5 years following diagnosis. Many studies have focused on the genetics of GBM with the aim of improving the prognosis of GBM patients. We investigated specific genes whose expressions are significantly related to both the length of the overall survival and the progression-free survival in patients with GBM. We obtained data for 12,042 gene mRNA expressions in 525 GBM tissues from the Cancer Genome Atlas (TCGA) database. Among those genes, we identified independent genes significantly associated with the prognosis of GBM. Receiver operating characteristic (ROC) curve analysis was performed to determine the genes significant for predicting the long-term survival of patients with GBM. Bioinformatics analysis was also performed for the significant genes. We identified 33 independent genes whose expressions were significantly associated with the prognosis of 525 patients with GBM. Among them, the expressions of five genes were independently associated with an improved prognosis of GBM, and the expressions of 28 genes were independently related to a poorer prognosis of GBM. The expressions of the ADAM22, ATP5C1, RAC3, SHANK1, AEBP1, C1RL, CHL1, CHST2, EFEMP2, and PGCP genes were either positively or negatively related to the long-term survival of GBM patients. Using a large-scale and open database, we found genes significantly associated with both the prognosis and long-term survival of patients with GBM. We believe that our findings may contribute to improving the understanding of the mechanisms underlying GBM.

A central question for the cancer research community is how to use genomic and proteomic data to guide cancer care. We investigated two potential clinical applications of The Cancer Genome Atlas (TCGA) data: predicting patient prognosis and identifying clinically actionable alterations. For four TCGA cohorts, we systematically evaluated the power of molecular data with or without clinical variables in predicting patient survival; and to facilitate further community efforts, we established an open-access model evaluation platform. We demonstrate that incorporating molecular data with clinical variables can improve predictive power. Interestingly, one model derived from ovarian cancer can predict outcomes in kidney cancer, suggesting the potential utility of cross-tumor analysis. Across 12 cancer types, we identified 10,281 somatic alterations in clinically relevant genes in 2,928 out of 3,277 patients (89.4%), revealing recurrent and potentially targetable alterations not revealed in single-tumor datasets. Our study represents the first systematic and comprehensive analysis evaluating the prognostic power of different types of molecular data across multiple cancer types. In addition, the clinically actionable alterations we identified may inform novel clinical trial design and treatment choice. Citation Format: Han Liang, TCGA Pan-Cancer Clinical/Predictor Group. Dissect the clinical utility of TCGA genomic data across tumor types. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 4168. doi:10.1158/1538-7445.AM2014-4168

Although the range of immune responses to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is variable, cytokine storm is observed in a subset of symptomatic individuals. To further understand the disease pathogenesis and, consequently, to develop an additional tool for clinicians to evaluate patients for presumptive intervention, we sought to compare plasma cytokine levels between a range of donor and patient samples grouped by a COVID-19 Severity Score (CSS) based on the need for hospitalization and oxygen requirement. Here we utilize a mutual information algorithm that classifies the information gain for CSS prediction provided by cytokine expression levels and clinical variables. Using this methodology, we found that a small number of clinical and cytokine expression variables are predictive of presenting COVID-19 disease severity, raising questions about the mechanism by which COVID-19 creates severe illness. The variables that were the most predictive of CSS included clinical variables such as age and abnormal chest x-ray as well as cytokines such as macrophage colony-stimulating factor, interferon-inducible protein 10, and interleukin-1 receptor antagonist. Our results suggest that SARS-CoV-2 infection causes a plethora of changes in cytokine profiles and that particularly in severely ill patients, these changes are consistent with the presence of macrophage activation syndrome and could furthermore be used as a biomarker to predict disease severity.

Glioblastoma (GBM) is the most common primary intracranial tumor with extremely high malignancy and poor prognosis. In order to identify the GBM prognostic biomarkers and establish a prognostic model, we analyzed the expression profile data of GBM in The Cancer Genome Atlas (TCGA) database as the experimental group. First, we identified the differentially expressed genes of different survival periods among the GBM patients. The GISTIC software and Kaplan Meier (KM) survival curve were used to analyze the copy number variation of GBM to identify the survival-associated amplified gene (SAG). We selected the intersection genes of up-regulated ones in short survival group and SAG, performed univariate Cox regression and iterative Lasso regression with them to identify the important candidate genes and establish a prognostic model. Based on the model, the prognostic score was calculated. The patients were divided into high-risk and low-risk groups according to the median prognostic score. Meanwhile ROC curve was used to evaluate the validity of the model, applying the KM survival analysis of the high-risk and low-risk groups. Multivariate Cox regression analysis was used to determine the independence of the prognostic score. All the data were verified with three external datasets: GEO GSE16011, CGGA, and Rembrandt. The results showed that differential expression analysis of different survival periods of GBM identified 426 up-regulated genes and 65 down-regulated genes in the TCGA GBM dataset. The intersection of up-regulated genes in short survival group and SAG yielded 47 genes. After the screening, the six-gene combination (EN2,PPBP,LRRC61,SEL1L3,CPA4,DDIT4L) prognostic model was finally determined. The area under ROC curve of the model in TCGA experimental group and three external validation group were all greater than 0.6, even reaching 0.912. KM analysis showed that the prognosis of the high-risk and low-risk groups was significant different (P<0.05). In the multivariate Cox regression analysis, the six-gene prognostic score was an independent factor influencing the prognosis of GBM patients (P<0.05). In summary, this study established a prognostic model of six-gene (EN2,PPBP,LRRC61,SEL1L3,CPA4,DDIT4L) for GBM. This six-gene model has good predictive ability and could be used as an independent prognostic marker for GBM patients.

Artificial neural network to classify cognitive impairment using gait and clinical variables

Somatic copy number alterations are associated with EGFR amplification and shortened survival in patients with primary glioblastoma

Background: Studies by The Cancer Genome Atlas (TCGA) and others have identified regions of somatic copy number alteration (SCNA) in cervical squamous cell carcinoma (CESC), head and neck squamous cell carcinoma (HNSC), and lung squamous cell carcinoma (LUSC). These three tumor types exhibit frequent SCNAs in chromosome 11, including focal gains in 11q13 and 11q22, as well as broad losses in 11p and 11q22-qter. Because expression levels are affected by underlying genomic events, we hypothesized that an integrated analysis of multiple genomic data types would provide increased ability to identify target genes in SCNA regions when compared to methods based on copy number alone while providing insight into mechanisms regulating expression. Examination of multiple tumor types provides a method to validate findings. Techniques: Gene expression (GE), DNA copy number (CN), DNA methylation (ME), and microRNA expression (miR) data were obtained from the TCGA studies of CESC, HNSC, and LUSC. For each tumor type, univariate and multivariate linear models were constructed on a gene-by-gene basis to investigate the effect of changes in CN, ME, and miR on GE. Analysis of model output provided an approach to identify target genes in SCNA regions and assess the effect of genomic alterations on expression. Results: Genome-wide GE, CN, ME, and miR data were available for n = 191 (CESC), n = 453 (HNSC), and n = 326 (LUSC) tumor samples. Univariate modeling detected a strong overall association between GE and CN, as measured by the coefficient of determination (model R2). Underlying CN events produced high model R2 for widely recognized genes, e.g. FADD in HNSC and LUSC, and less well-known genes. For example, DCUN1D5 (11q22) exhibits high model R2 in all three tumor types. Univariate analyses suggest that ALDH3B1 (11q13), a predicted target of the tumor suppressor miR-205, is epigenetically regulated. Most samples exhibit high expression of miR-205 and moderate expression of ALDH3B1, but increased ALDH3B1 levels are seen when miR-205 levels are low. This is notable in light of the striking negative association between miR-205 expression and methylation levels, particularly in CESC and LUSC. Fanconi anemia family member FANCF lies in a broad region of CN loss in 11p that contains no clear driver gene. Whereas univariate associations between FANCF GE and CN were moderate, multivariate analyses highlighted this gene because of the additional effect of ME on GE, which was pronounced in CESC and LUSC. Conclusion: Linear modeling techniques provide a flexible and powerful basis for performing integrated analysis of genomic data. Our approach produces predicted results when analyzing known cancer genes, highlights lesser known genes for future study, provides insight into gene regulation, and draws attention to genes relevant in multiple tumor types. Citation Format: Vonn Walter, Ying Du, Xiaoying Yin, Wei Sun, Matthew D. Wilkerson, Michele C. Hayward, Ashley H. Salazar, Charles M. Perou, David N. Hayes. Integrated genomic analysis of chromosome 11 in squamous tumors. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 4801. doi:10.1158/1538-7445.AM2015-4801

ObjectiveIn the development of immunotherapies in gliomas, the tumor microenvironment (TME) needs to be investigated. We aimed to construct a prognostic microenvironment-related immune signature via ESTIMATE (PROMISE model) for glioma.MethodsStromal score (SS) and immune score (IS) were calculated via ESTIMATE for each glioma sample in the cancer genome atlas (TCGA), and differentially expressed genes (DEGs) were identified between high-score and low-score groups. Prognostic DEGs were selected via univariate Cox regression analysis. Using the lower-grcade glioma (LGG) data set in TCGA, we performed LASSO regression based on the prognostic DEGs and constructed a PROMISE model for glioma. The model was validated with survival analysis and the receiver operating characteristic (ROC) in TCGA glioma data sets (LGG, glioblastoma multiforme [GBM] and LGG+GBM) and Chinese glioma genome atlas (CGGA). A nomogram was developed to predict individual survival chances. Further, we explored the underlying mechanisms using gene set enrichment analysis (GSEA) and Cibersort analysis of tumor-infiltrating immune cells between risk groups as defined by the PROMISE model.ResultsWe obtained 220 upregulated DEGs and 42 downregulated DEGs in both high-IS and high-SS groups. The Cox regression highlighted 155 prognostic DEGs, out of which we selected 4 genes (CD86, ANXA1, C5AR1, and CD5) to construct a PROMISE model. The model stratifies glioma patients in TCGA as well as in CGGA with distinct survival outcome (P<0.05, Hazard ratio [HR]>1) and acceptable predictive accuracy (AUCs>0.6). With the nomogram, an individualized survival chance could be predicted intuitively with specific age, tumor grade, Isocitrate dehydrogenase (IDH) status, and the PROMISE risk score. ROC showed significant discrimination with the area under curves (AUCs) of 0.917 and 0.817 in TCGA and CGGA, respectively. GSEA between risk groups in both data sets were significantly enriched in multiple immune-related pathways. The Cibersort analysis highlighted four immune cells, i.e., CD 8 T cells, neutrophils, follicular helper T (Tfh) cells, and Natural killer (NK) cells.ConclusionsThe PROMISE model can further stratify both LGG and GBM patients with distinct survival outcomes.These findings may help further our understanding of TME in gliomas and shed light on immunotherapies.

Pathologic review of tumor morphology in histologic sections is the traditional method for cancer classification and grading, yet human review has limitations that can result in low reproducibility and inter-observer agreement. Computerized image analysis can partially overcome these shortcomings due to its capacity to quantitatively and reproducibly measure histologic structures on a large-scale. In this paper, we present an end-to-end image analysis and data integration pipeline for large-scale morphologic analysis of pathology images and demonstrate the ability to correlate phenotypic groups with molecular data and clinical outcomes. We demonstrate our method in the context of glioblastoma (GBM), with specific focus on the degree of the oligodendroglioma component. Over 200 million nuclei in digitized pathology slides from 117 GBMs in the Cancer Genome Atlas were quantitatively analyzed, followed by multiplatform correlation of nuclear features with molecular and clinical data. For each nucleus, a Nuclear Score (NS) was calculated based on the degree of oligodendroglioma appearance, using a regression model trained from the optimal feature set. Using the frequencies of neoplastic nuclei in low and high NS intervals, we were able to cluster patients into three well-separated disease groups that contained low, medium, or high Oligodendroglioma Component (OC). We showed that machine-based classification of GBMs with high oligodendroglioma component uncovered a set of tumors with strong associations with PDGFRA amplification, proneural transcriptional class, and expression of the oligodendrocyte signature genes MBP, HOXD1, PLP1, MOBP and PDGFRA. Quantitative morphologic features within the GBMs that correlated most strongly with oligodendrocyte gene expression were high nuclear circularity and low eccentricity. These findings highlight the potential of high throughput morphologic analysis to complement and inform human-based pathologic review.

Purpose: Alternative splicing (AS) was reported to play a vital role in development and progression of glioblastoma (GBM), the most common and fatal brain tumor. Systematic analysis of survival-associated AS event profiles and prognostic prediction model based on multiple AS events in GBM was needed.Methods: Genome-wide AS and RNA sequencing profiles were generated in 152 patients with GBM in the cancer genome atlas (TCGA). Prognosis-associated AS events were screened by integrated Cox regression analysis to construct the prognostic risk score model in the training cohort (n = 101). The AS-based signature and clinicopathologic parameters were applied to construct a prognostic nomogram for 0.5-, 1-, and 3-year OS prediction. Finally, the regulatory networks between prognostic AS events and splicing factors (SFs) were constructed.Results: A total of 1,598 prognosis-related AS events from 1,183 source genes were determined. Eight prognostic risk score model based on integrated AS events and 7 AS types were established, respectively. Concordance index (C-index) and receiver operating characteristic (ROC) curve analysis demonstrated powerful ability in distinguishing patients' outcomes. Only Alternate Donor site (AD) and Exon Skip (ES) signature out of the eight types of AS signature were identified as independent prognostic factors for GBM, which was validated in the internal validation cohort. The nomogram with age, new event, pharmaceutical therapy, radiation therapy, AD signature and ES signature were constructed, with C-index of 0.892 (95% CI, 0.853–0.931; P = 5.13 × 10−15). Calibration plots, ROC, and decision curve analysis suggested excellent predictive performance for the nomogram in both TCGA training cohort and validation cohort. Splicing network indicated distinguished correlations between prognostic AS events and SFs in GBM patients.Conclusions: AS-based prediction model could serve as a promising prognostic predictor and potential therapeutic target for GBM, facilitating better treatment strategies in clinical practice.

Neuromedin B (NMB) is a neuropeptide that plays a key role in many physiological processes and is involved in the pathology of various diseases. Increased levels of NMB have been reported in solid tumors. Therefore, we investigated the prognostic value of NMB in glioblastoma (GBM). Expression profiles of NMB mRNA were investigated in GBM and normal tissues using data from the cancer genome atlas (TCGA). NMB protein expression was obtained using data from the Human Protein Atlas. Receiver operating characteristic (ROC) curves were evaluated in GBM and normal tissues. The survival effect of NMB in GBM patients was evaluated using the Kaplan-Meier method. Protein-protein interaction networks were constructed using STRING, and the functional enrichment analyses were performed. The relationship between NMB expression and tumor-infiltrating lymphocytes was analyzed using the Tumor Immune Estimation Resource (TIMER) and the Tumor-Immune System Interaction database (TISIDB). NMB was overexpressed in GBM relative to normal biopsy specimens. The ROC analysis showed that the sensitivity and specificity of NMB in GBM were 96.4% and 96.2%, respectively. Kaplan-Meier survival analysis showed that GBM patients with high NMB expression had a better prognosis than those with low NMB expression (16.3 vs. 12.7 months, p = 0.002). Correlation analysis showed that NMB expression was associated with tumor-infiltrating lymphocytes and tumor purity. High expression of NMB was associated with increased GBM patient survival. Our study indicated that the NMB expression may be a biomarker for prognosis and that NMB may be an immunotherapy target in GBM.