Genomic deletion of malic enzyme 2 confers collateral lethality in pancreatic cancer.

The genome of pancreatic ductal adenocarcinoma (PDAC) frequently contains deletions of tumour suppressor gene loci, most notably SMAD4, which is homozygously deleted in nearly one-third of cases. As loss of neighbouring housekeeping genes can confer collateral lethality, we sought to determine whether loss of the metabolic gene malic enzyme 2 (ME2) in the SMAD4 locus would create cancer-specific metabolic vulnerability upon targeting of its paralogous isoform ME3. The mitochondrial malic enzymes (ME2 and ME3) are oxidative decarboxylases that catalyse the conversion of malate to pyruvate and are essential for NADPH regeneration and reactive oxygen species homeostasis. Here we show that ME3 depletion selectively kills ME2-null PDAC cells in a manner consistent with an essential function for ME3 in ME2-null cancer cells. Mechanistically, integrated metabolomic and molecular investigation of cells deficient in mitochondrial malic enzymes revealed diminished NADPH production and consequent high levels of reactive oxygen species. These changes activate AMP activated protein kinase (AMPK), which in turn directly suppresses sterol regulatory element-binding protein 1 (SREBP1)-directed transcription of its direct targets including the BCAT2 branched-chain amino acid transaminase 2) gene. BCAT2 catalyses the transfer of the amino group from branched-chain amino acids to α-ketoglutarate (α-KG) thereby regenerating glutamate, which functions in part to support de novo nucleotide synthesis. Thus, mitochondrial malic enzyme deficiency, which results in impaired NADPH production, provides a prime 'collateral lethality' therapeutic strategy for the treatment of a substantial fraction of patients diagnosed with this intractable disease.

Synthetic vulnerabilities of mesenchymal subpopulations in pancreatic cancer.

Malignant neoplasms evolve in response to changes in oncogenic signalling. Cancer cell plasticity in response to evolutionary pressures is fundamental to tumour progression and the development of therapeutic resistance. Here we determine the molecular and cellular mechanisms of cancer cell plasticity in a conditional oncogenic Kras mouse model of pancreatic ductal adenocarcinoma (PDAC), a malignancy that displays considerable phenotypic diversity and morphological heterogeneity. In this model, stochastic extinction of oncogenic Kras signalling and emergence of Kras-independent escaper populations (cells that acquire oncogenic properties) are associated with de-differentiation and aggressive biological behaviour. Transcriptomic and functional analyses of Kras-independent escapers reveal the presence of Smarcb1-Myc-network-driven mesenchymal reprogramming and independence from MAPK signalling. A somatic mosaic model of PDAC, which allows time-restricted perturbation of cell fate, shows that depletion of Smarcb1 activates the Myc network, driving an anabolic switch that increases protein metabolism and adaptive activation of endoplasmic-reticulum-stress-induced survival pathways. Increased protein turnover renders mesenchymal sub-populations highly susceptible to pharmacological and genetic perturbation of the cellular proteostatic machinery and the IRE1-α-MKK4 arm of the endoplasmic-reticulum-stress-response pathway. Specifically, combination regimens that impair the unfolded protein responses block the emergence of aggressive mesenchymal subpopulations in mouse and patient-derived PDAC models. These molecular and biological insights inform a potential therapeutic strategy for targeting aggressive mesenchymal features of PDAC.

Enhancer reprogramming in PRC2-deficient malignant peripheral nerve sheath tumors induces a targetable de-differentiated state.

Malignant peripheral nerve sheath tumors (MPNSTs) are soft tissue sarcomas that frequently harbor genetic alterations in polycomb repressor complex 2 (PRC2) components-SUZ12 and EED. Here, we show that PRC2 loss confers a dedifferentiated early neural-crest phenotype which is exclusive to PRC2-mutant MPNSTs and not a feature of neurofibromas. Neural crest phenotype in PRC2 mutant MPNSTs was validated via cross-species comparative analysis using spontaneous and transgenic MPNST models. Systematic chromatin state profiling of the MPNST cells showed extensive epigenomic reprogramming or chromatin states associated with PRC2 loss and identified gains of active enhancer states/super-enhancers on early neural crest regulators in PRC2-mutant conditions around genomic loci that harbored repressed/poised states in PRC2-WT MPNST cells. Consistently, inverse correlation between H3K27me3 loss and H3K27Ac gain was noted in MPNSTs. Epigenetic editing experiments established functional roles for enhancer gains on DLX5-a key regulator of neural crest phenotype. Consistently, blockade of enhancer activity by bromodomain inhibitors specifically suppressed this neural crest phenotype and tumor burden in PRC2-mutant PDXs. Together, these findings reveal accumulation of dedifferentiated neural crest like state in PRC2-mutant MPNSTs that can be targeted by enhancer blockade.

Epithelial-to-mesenchymal transition is dispensable for metastasis but induces chemoresistance in pancreatic cancer.

Diagnosis of pancreatic ductal adenocarcinoma (PDAC) is associated with a dismal prognosis despite current best therapies; therefore new treatment strategies are urgently required. Numerous studies have suggested that epithelial-to-mesenchymal transition (EMT) contributes to early-stage dissemination of cancer cells and is pivotal for invasion and metastasis of PDAC. EMT is associated with phenotypic conversion of epithelial cells into mesenchymal-like cells in cell culture conditions, although such defined mesenchymal conversion (with spindle-shaped morphology) of epithelial cells in vivo is rare, with quasi-mesenchymal phenotypes occasionally observed in the tumour (partial EMT). Most studies exploring the functional role of EMT in tumours have depended on cell-culture-induced loss-of-function and gain-of-function experiments involving EMT-inducing transcription factors such as Twist, Snail and Zeb1 (refs 2, 3, 7-10). Therefore, the functional contribution of EMT to invasion and metastasis remains unclear, and genetically engineered mouse models to address a causal connection are lacking. Here we functionally probe the role of EMT in PDAC by generating mouse models of PDAC with deletion of Snail or Twist, two key transcription factors responsible for EMT. EMT suppression in the primary tumour does not alter the emergence of invasive PDAC, systemic dissemination or metastasis. Suppression of EMT leads to an increase in cancer cell proliferation with enhanced expression of nucleoside transporters in tumours, contributing to enhanced sensitivity to gemcitabine treatment and increased overall survival of mice. Collectively, our study suggests that Snail- or Twist-induced EMT is not rate-limiting for invasion and metastasis, but highlights the importance of combining EMT inhibition with chemotherapy for the treatment of pancreatic cancer.

Genomics and the Immune Landscape of Osteosarcoma.

Conventional osteosarcoma (OS) is a high-grade intraosseous malignancy with production of osteoid matrix; however, a deeper dive into the underlying genetics reveals genomic complexity and instability that result in significant tumor heterogeneity. While early karyotyping studies demonstrated aneuploidy with chromosomal complexity and structural rearrangements, further investigations have identified few recurrent genetic alterations with the exception of the tumor suppressors TP53 and RB1. More recent studies utilizing next-generation sequencing (NGS; whole-exome sequencing, WES; and whole-genome sequencing, WGS) reveal a genomic landscape predominantly characterized by somatic copy number alterations rather than point/indel mutations. Despite its genomic complexity, OS has shown variable immune infiltrate and limited immunogenicity. In the current chapter, we review the hallmarks of OS genomics across recent NGS studies and the immune profile of OS including a large institutional cohort of OS patients with recurrent and metastatic disease. Understanding the genomic and immune landscape of OS may provide opportunities for translation in both molecularly targeted therapies and novel immuno-oncology approaches.

FusionPathway: Prediction of pathways and therapeutic targets associated with gene fusions in cancer.

Numerous gene fusions have been uncovered across multiple cancer types. Although the ability to target several of these fusions has led to the development of some successful anti-cancer drugs, most of them are not druggable. Understanding the molecular pathways of a fusion is important in determining its function in oncogenesis and in developing therapeutic strategies for patients harboring the fusion. However, the molecular pathways have been elucidated for only a few fusions, in part because of the labor-intensive nature of the required functional assays. Therefore, we developed a domain-based network approach to infer the pathways of a fusion. Molecular interactions of a fusion are first predicted by using its protein domain composition, and its associated pathways are then inferred from these molecular interactions. We demonstrated the capabilities of this approach by primarily applying it to the well-studied BCR-ABL1 fusion. The approach was also applied to two undruggable fusions in sarcoma, EWS-FL1 and FUS-DDIT3. We successfully identified known genes and pathways associated with these fusions and satisfactorily validated these predictions using several benchmark sets. The predictions of EWS-FL1 and FUS-DDIT3 also correlate with results of high-throughput drug screening. To our best knowledge, this is the first approach for inferring pathways of fusions.

Truncating PREX2 mutations activate its GEF activity and alter gene expression regulation in NRAS-mutant melanoma.

PREX2 (phosphatidylinositol-3,4,5-triphosphate-dependent Rac-exchange factor 2) is a PTEN (phosphatase and tensin homolog deleted on chromosome 10) binding protein that is significantly mutated in cutaneous melanoma and pancreatic ductal adenocarcinoma. Here, genetic and biochemical analyses were conducted to elucidate the nature and mechanistic basis of PREX2 mutation in melanoma development. By generating an inducible transgenic mouse model we showed an oncogenic role for a truncating PREX2 mutation (PREX2(E824)*) in vivo in the context of mutant NRAS. Using integrative cross-species gene expression analysis, we identified deregulated cell cycle and cytoskeleton organization as significantly perturbed biological pathways in PREX2 mutant tumors. Mechanistically, truncation of PREX2 activated its Rac1 guanine nucleotide exchange factor activity, abolished binding to PTEN and activated the PI3K (phosphatidyl inositol 3 kinase)/Akt signaling pathway. We further showed that PREX2 truncating mutations or PTEN deletion induces down-regulation of the tumor suppressor and cell cycle regulator CDKN1C (also known as p57(KIP2)). This down-regulation occurs, at least partially, through DNA hypomethylation of a differentially methylated region in chromosome 11 that is a known regulatory region for expression of the CDKN1C gene. Together, these findings identify PREX2 as a mediator of NRAS-mutant melanoma development that acts through the PI3K/PTEN/Akt pathway to regulate gene expression of a cell cycle regulator.

Genes suppressed by DNA methylation in non-small cell lung cancer reveal the epigenetics of epithelial-mesenchymal transition.

BACKGROUND: DNA methylation is associated with aberrant gene expression in cancer, and has been shown to correlate with therapeutic response and disease prognosis in some types of cancer. We sought to investigate the biological significance of DNA methylation in lung cancer. RESULTS: We integrated the gene expression profiles and data of gene promoter methylation for a large panel of non-small cell lung cancer cell lines, and identified 578 candidate genes with expression levels that were inversely correlated to the degree of DNA methylation. We found these candidate genes to be differentially methylated in normal lung tissue versus non-small cell lung cancer tumors, and segregated by histologic and tumor subtypes. We used gene set enrichment analysis of the genes ranked by the degree of correlation between gene expression and DNA methylation to identify gene sets involved in cellular migration and metastasis. Our unsupervised hierarchical clustering of the candidate genes segregated cell lines according to the epithelial-to-mesenchymal transition phenotype. Genes related to the epithelial-to-mesenchymal transition, such as AXL, ESRP1, HoxB4, and SPINT1/2, were among the nearly 20% of the candidate genes that were differentially methylated between epithelial and mesenchymal cells. Greater numbers of genes were methylated in the mesenchymal cells and their expressions were upregulated by 5-azacytidine treatment. Methylation of the candidate genes was associated with erlotinib resistance in wild-type EGFR cell lines. The expression profiles of the candidate genes were associated with 8-week disease control in patients with wild-type EGFR who had unresectable non-small cell lung cancer treated with erlotinib, but not in patients treated with sorafenib. CONCLUSIONS: Our results demonstrate that the underlying biology of genes regulated by DNA methylation may have predictive value in lung cancer that can be exploited therapeutically.

Identification of cancer fusion drivers using network fusion centrality.

SUMMARY: Gene fusions are being discovered at an increasing rate using massively parallel sequencing technologies. Prioritization of cancer fusion drivers for validation cannot be performed using traditional single-gene based methods because fusions involve portions of two partner genes. To address this problem, we propose a novel network analysis method called fusion centrality that is specifically tailored for prioritizing gene fusions. We first propose a domain-based fusion model built on the theory of exon/domain shuffling. The model leads to a hypothesis that a fusion is more likely to be an oncogenic driver if its partner genes act like hubs in a network because the fusion mutation can deregulate normal functions of many other genes and their pathways. The hypothesis is supported by the observation that for most known cancer fusion genes, at least one of the fusion partners appears to be a hub in a network, and even for many fusions both partners appear to be hubs. Based on this model, we construct fusion centrality, a multi-gene-based network metric, and use it to score fusion drivers. We show that the fusion centrality outperforms other single gene-based methods. Specifically, the method successfully predicts most of 38 newly discovered fusions that had validated oncogenic importance. To our best knowledge, this is the first network-based approach for identifying fusion drivers. AVAILABILITY: Matlab code implementing the fusion centrality method is available upon request from the corresponding authors.

Targeting IL-11R/EZH2 signaling axis as a therapeutic strategy for osteosarcoma lung metastases.

Lung metastases are the primary cause of death for osteosarcoma (OS) patients. We recently validated interleukin-11 receptor α (IL-11Rα) as a molecular target for the inhibition of OS lung metastases. Since there is no clinically approved antibody against this receptor, we sought to identify downstream targets that mediate the effects of IL-11Rα signaling. We used shRNA to deplete IL-11Rα from OS cells; as a complementary approach, we added IL-11 exogenously to OS cells. The resulting changes in gene expression identified EZH2 as a downstream candidate. This was confirmed by knockdown of IL-11Rα in OS cells, which led to increased expression of genes repressed by histone methyltransferase EZH2, including members of the WNT pathway, a known target pathway of EZH2. Exogenous IL-11 increased the global levels of histone H3 lysine 27 trimethylation, evidence of EZH2 activation. Treatment with the EZH2 inhibitor GSK126 significantly reduced in vitro proliferation and increased cell-cycle arrest and apoptosis, which were partially mediated through the WNT pathway. In vivo, treatment of an orthotopic nude mouse model of OS with GSK126 inhibited lung metastatic growth and prolonged survival. In addition, significantly shorter recurrence-free survival was seen in OS patients with high levels of EZH2 in their primary tumors (P < .05). This suggests that IL-11Rα promotes OS lung metastasis via activation of EZH2. Thus, blocking EZH2 activity may be an effective strategy for inhibiting OS lung metastasis and improving prognosis.

Telomere dysfunction alters intestinal stem cell dynamics to promote cancer.

Telomere dynamics are linked to aging hallmarks, and age-associated telomere loss fuels the development of epithelial cancers. In Apc-mutant mice, the onset of DNA damage associated with telomere dysfunction has been shown to accelerate adenoma initiation via unknown mechanisms. Here, we observed that Apc-mutant mice engineered to experience telomere dysfunction show accelerated adenoma formation resulting from augmented cell competition and clonal expansion. Mechanistically, telomere dysfunction induces the repression of EZH2, resulting in the derepression of Wnt antagonists, which causes the differentiation of adjacent stem cells and a relative growth advantage to Apc-deficient telomere dysfunctional cells. Correspondingly, in this mouse model, GSK3ß inhibition countered the actions of Wnt antagonists on intestinal stem cells, resulting in impaired adenoma formation of telomere dysfunctional Apc-mutant cells. Thus, telomere dysfunction contributes to cancer initiation through altered stem cell dynamics, identifying an interception strategy for human APC-mutant cancers with shortened telomeres.

Loss of the DNA Repair Gene RNase H2 Identifies a Unique Subset of DDR-Deficient Leiomyosarcomas.

Targeting the DNA damage response (DDR) pathway is an emerging therapeutic approach for leiomyosarcoma (LMS), and loss of RNase H2, a DDR pathway member, is a potentially actionable alteration for DDR-targeted treatments. Therefore, we designed a protein- and genomic-based RNase H2 screening assay to determine its prevalence and prognostic significance. Using a selective RNase H2 antibody on a pan-tumor microarray (TMA), RNase H2 loss was more common in LMS (11.5%, 9/78) than across all tumors (3.8%, 32/843). In a separate LMS cohort, RNase H2 deficiency was confirmed in uterine LMS (U-LMS, 21%, 23/108) and soft-tissue LMS (ST-LMS; 30%, 39/102). In the TCGA database, RNASEH2B homozygous deletions (HomDels) were found in 6% (5/80) of LMS cases, with a higher proportion in U-LMS (15%; 4/27) compared with ST-LMS (2%; 1/53). Using the SNiPDx targeted-NGS sequencing assay to detect biallelic loss of function in select DDR-related genes, we found RNASEH2B HomDels in 54% (19/35) of U-LMS cases with RNase H2 loss by IHC, and 7% (3/43) HomDels in RNase H2 intact cases. No RNASEH2B HomDels were detected in ST-LMS. In U-LMS patient cohort (n = 109), no significant overall survival difference was seen in patients with RNase H2 loss versus intact, or RNASEH2B HomDel (n = 12) versus Non-HomDel (n = 37). The overall diagnostic accuracy, sensitivity, and specificity of RNase H2 IHC for detecting RNA-SEH2B HomDels in U-LMS was 76%, 93%, and 71%, respectively, and it is being developed for future predictive biomarker driven clinical trials targeting DDR in U-LMS.

A randomized, non-comparative phase 2 study of neoadjuvant immune-checkpoint blockade in retroperitoneal dedifferentiated liposarcoma and extremity/truncal undifferentiated pleomorphic sarcoma.

Based on the demonstrated clinical activity of immune-checkpoint blockade (ICB) in advanced dedifferentiated liposarcoma (DDLPS) and undifferentiated pleomorphic sarcoma (UPS), we conducted a randomized, non-comparative phase 2 trial ( NCT03307616 ) of neoadjuvant nivolumab or nivolumab/ipilimumab in patients with resectable retroperitoneal DDLPS (n = 17) and extremity/truncal UPS (+ concurrent nivolumab/radiation therapy; n = 10). The primary end point of pathologic response (percent hyalinization) was a median of 8.8% in DDLPS and 89% in UPS. Secondary end points were the changes in immune infiltrate, radiographic response, 12- and 24-month relapse-free survival and overall survival. Lower densities of regulatory T cells before treatment were associated with a major pathologic response (hyalinization > 30%). Tumor infiltration by B cells was increased following neoadjuvant treatment and was associated with overall survival in DDLPS. B cell infiltration was associated with higher densities of regulatory T cells before treatment, which was lost upon ICB treatment. Our data demonstrate that neoadjuvant ICB is associated with complex immune changes within the tumor microenvironment in DDLPS and UPS and that neoadjuvant ICB with concurrent radiotherapy has significant efficacy in UPS.

Complete loss of TP53 and RB1 is associated with complex genome and low immune infiltrate in pleomorphic rhabdomyosarcoma.

Rhabdomyosarcoma accounts for roughly 1% of adult sarcomas, with pleomorphic rhabdomyosarcoma (PRMS) as the most common subtype. Survival outcomes remain poor for patients with PRMS, and little is known about the molecular drivers of this disease. To better characterize PRMS, we performed a broad array of genomic and immunostaining analyses on 25 patient samples. In terms of gene expression and methylation, PRMS clustered more closely with other complex karyotype sarcomas than with pediatric alveolar and embryonal rhabdomyosarcoma. Immune infiltrate levels in PRMS were among the highest observed in multiple sarcoma types and contrasted with low levels in other rhabdomyosarcoma subtypes. Lower immune infiltrate was associated with complete loss of both TP53 and RB1. This comprehensive characterization of the genetic, epigenetic, and immune landscape of PRMS provides a roadmap for improved prognostications and therapeutic exploration.

Prediction of biomarkers and therapeutic combinations for anti-PD-1 immunotherapy using the global gene network association.

Owing to a lack of response to the anti-PD1 therapy for most cancer patients, we develop a network approach to infer genes, pathways, and potential therapeutic combinations that are associated with tumor response to anti-PD1. Here, our prediction identifies genes and pathways known to be associated with anti-PD1, and is further validated by 6 CRISPR gene sets associated with tumor resistance to cytotoxic T cells and targets of the 36 compounds that have been tested in clinical trials for combination treatments with anti-PD1. Integration of our top prediction and TCGA data identifies hundreds of genes whose expression and genetic alterations that could affect response to anti-PD1 in each TCGA cancer type, and the comparison of these genes across cancer types reveals that the tumor immunoregulation associated with response to anti-PD1 would be tissue-specific. In addition, the integration identifies the gene signature to calculate the MHC I association immunoscore (MIAS) that shows a good correlation with patient response to anti-PD1 for 411 melanoma samples complied from 6 cohorts. Furthermore, mapping drug target data to the top genes in our association prediction identifies inhibitors that could potentially enhance tumor response to anti-PD1, such as inhibitors of the encoded proteins of CDK4, GSK3B, and PTK2.

Identification of Functional Heterogeneity of Carcinoma-Associated Fibroblasts with Distinct IL6-Mediated Therapy Resistance in Pancreatic Cancer.

The tumor microenvironment in pancreatic ductal adenocarcinoma (PDAC) involves a significant accumulation of fibroblasts as part of the host response to cancer. Using single-cell RNA sequencing, multiplex immunostaining, and several genetic mouse models, we identify carcinoma-associated fibroblasts (CAF) with opposing functions in PDAC progression. Depletion of fibroblast activation protein (FAP)+ CAFs results in increased survival, in contrast to depletion of alpha smooth muscle actin (αSMA)+ CAFs, which leads to decreased survival. Tumor-promoting FAP+ CAFs (TP-CAF) and tumor-restraining αSMA+ CAFs (TR-CAF) differentially regulate cancer-associated pathways and accumulation of regulatory T cells. Improved efficacy of gemcitabine is observed when IL6 is deleted from αSMA+ CAFs but not from FAP+ CAFs using dual-recombinase genetic PDAC models. Improved gemcitabine efficacy due to lack of IL6 synergizes with anti-PD-1 immunotherapy to significantly improve survival of PDAC mice. Our study identifies functional heterogeneity of CAFs in PDAC progression and their different roles in therapy response. SIGNIFICANCE: PDAC is associated with accumulation of dense stroma consisting of fibroblasts and extracellular matrix that regulate tumor progression. Here, we identify two distinct populations of fibroblasts with opposing roles in the progression and immune landscape of PDAC. Our findings demonstrate that fibroblasts are functionally diverse with therapeutic implications. This article is highlighted in the In This Issue feature, p. 1397.

Multi-site desmoplastic small round cell tumors are genetically related and immune-cold.

Desmoplastic small round cell tumor (DSRCT) is a highly aggressive soft tissue sarcoma that is characterized by the EWSR1-WT1 fusion protein. Patients present with hundreds of tumor implants in their abdominal cavity at various sites. To determine the genetic relatedness among these sites, exome and RNA sequencing were performed on 22 DSRCT specimens from 14 patients, four of whom had specimens from various tissue sites. Multi-site tumors from individual DSRCT patients had a shared origin and were highly related. Other than the EWSR1-WT1 fusion, very few secondary cancer gene mutations were shared among the sites. Among these, ARID1A, was recurrently mutated, which corroborates findings by others in DSRCT patients. Knocking out ARID1A in JN-DSRCT cells using CRISPR/CAS9 resulted in significantly lower cell proliferation and increased drug sensitivity. The transcriptome data were integrated using network analysis and drug target database information to identify potential therapeutic opportunities in EWSR1-WT1-associated pathways, such as PI3K and mTOR pathways. Treatment of JN-DSRCT cells with the PI3K inhibitor alpelisib and mTOR inhibitor temsirolimus reduced cell proliferation. In addition, the low mutation burden was associated with an immune-cold state in DSRCT. Together, these data reveal multiple genomic and immune features of DSRCT and suggest therapeutic opportunities in patients.

The androgen receptor is a therapeutic target in desmoplastic small round cell sarcoma.

Desmoplastic small round cell tumor (DSRCT) is an aggressive, usually incurable sarcoma subtype that predominantly occurs in post-pubertal young males. Recent evidence suggests that the androgen receptor (AR) can promote tumor progression in DSRCTs. However, the mechanism of AR-induced oncogenic stimulation remains undetermined. Herein, we demonstrate that enzalutamide and AR-directed antisense oligonucleotides (AR-ASO) block 5α-dihydrotestosterone (DHT)-induced DSRCT cell proliferation and reduce xenograft tumor burden. Gene expression analysis and chromatin immunoprecipitation sequencing (ChIP-seq) were performed to elucidate how AR signaling regulates cellular epigenetic programs. Remarkably, ChIP-seq revealed novel DSRCT-specific AR DNA binding sites adjacent to key oncogenic regulators, including WT1 (the C-terminal partner of the pathognomonic fusion protein) and FOXF1. Additionally, AR occupied enhancer sites that regulate the Wnt pathway, neural differentiation, and embryonic organ development, implicating AR in dysfunctional cell lineage commitment. Our findings have direct clinical implications given the widespread availability of FDA-approved androgen-targeted agents used for prostate cancer.

Prediction of human functional genetic networks from heterogeneous data using RVM-based ensemble learning.

MOTIVATION: Three major problems confront the construction of a human genetic network from heterogeneous genomics data using kernel-based approaches: definition of a robust gold-standard negative set, large-scale learning and massive missing data values. RESULTS: The proposed graph-based approach generates a robust GSN for the training process of genetic network construction. The RVM-based ensemble model that combines AdaBoost and reduced-feature yields improved performance on large-scale learning problems with massive missing values in comparison to Naïve Bayes. CONTACT: dargenio@bmsr.usc.edu SUPPLEMENTARY INFORMATION: Supplementary material is available at Bioinformatics online.