Analysis and Optimization of Equitable US Cancer Clinical Trial Center Access by Travel Time

Importance

Racially minoritized and socioeconomically disadvantaged populations are currently underrepresented in clinical trials. Data-driven, quantitative analyses and strategies are required to help address this inequity.

Objective

To systematically analyze the geographical distribution of self-identified racial and socioeconomic demographics within commuting distance to cancer clinical trial centers and other hospitals in the US.

Design, Setting, and Participants

This longitudinal quantitative study used data from the US Census 2020 Decennial and American community survey (which collects data from all US residents), OpenStreetMap, National Cancer Institute-designated Cancer Centers list, Nature Index of Cancer Research Health Institutions, National Trial registry, and National Homeland Infrastructure Foundation-Level Data. Statistical analyses were performed on data collected between 2006 and 2020.

Main Outcomes and Measures

Population distributions of socioeconomic deprivation indices and self-identified race within 30-, 60-, and 120-minute 1-way driving commute times from US cancer trial sites. Map overlay of high deprivation index and high diversity areas with existing hospitals, existing major cancer trial centers, and commuting distance to the closest cancer trial center.

Results

The 78 major US cancer trial centers that are involved in 94% of all US cancer trials and included in this study were found to be located in areas with socioeconomically more affluent populations with higher proportions of self-identified White individuals (+10.1% unpaired mean difference; 95% CI, +6.8% to +13.7%) compared with the national average. The top 10th percentile of all US hospitals has catchment populations with a range of absolute sum difference from 2.4% to 35% from one-third each of Asian/multiracial/other (Asian alone, American Indian or Alaska Native alone, Native Hawaiian or Other Pacific Islander alone, some other race alone, population of 2 or more races), Black or African American, and White populations. Currently available data are sufficient to identify diverse census tracks within preset commuting times (30, 60, or 120 minutes) from all hospitals in the US (N = 7623). Maps are presented for each US city above 500 000 inhabitants, which display all prospective hospitals and major cancer trial sites within commutable distance to racially diverse and socioeconomically disadvantaged populations.

Conclusion and Relevance

This study identified biases in the sociodemographics of populations living within commuting distance to US-based cancer trial sites and enables the determination of more equitably commutable prospective satellite hospital sites that could be mobilized for enhanced racial and socioeconomic representation in clinical trials. The maps generated in this work may inform the design of future clinical trials or investigations in enrollment and retention strategies for clinical trials; however, other recruitment barriers still need to be addressed to ensure racial and socioeconomic demographics within the geographical vicinity of a clinical site can translate to equitable trial participant representation.

Mechanistically based blood proteomic markers in the TGF-β pathway stratify risk of hepatocellular cancer in patients with cirrhosis

Hepatocellular carcinoma (HCC) is the third leading cause of death from cancer worldwide but is often diagnosed at an advanced incurable stage. Yet, despite the urgent need for blood-based biomarkers for early detection, few studies capture ongoing biology to identify risk-stratifying biomarkers. We address this gap using the TGF-β pathway because of its biological role in liver disease and cancer, established through rigorous animal models and human studies. Using machine learning methods with blood levels of 108 proteomic markers in the TGF-β family, we found a pattern that differentiates HCC from non-HCC in a cohort of 216 patients with cirrhosis, which we refer to as TGF-β based Protein Markers for Early Detection of HCC (TPEARLE) comprising 31 markers. Notably, 20 of the patients with cirrhosis alone presented an HCC-like pattern, suggesting that they may be a group with as yet undetected HCC or at high risk for developing HCC. In addition, we found two other biologically relevant markers, Myostatin and Pyruvate Kinase M2 (PKM2), which were significantly associated with HCC. We tested these for risk stratification of HCC in multivariable models adjusted for demographic and clinical variables, as well as batch and site. These markers reflect ongoing biology in the liver. They potentially indicate the presence of HCC early in its evolution and before it is manifest as a detectable lesion, thereby providing a set of markers that may be able to stratify risk for HCC.

Genetic Ancestry Inference from Cancer-Derived Molecular Data across Genomic and Transcriptomic Platforms

Genetic ancestry-oriented cancer research requires the ability to perform accurate and robust genetic ancestry inference from existing cancer-derived data, including whole-exome sequencing, transcriptome sequencing, and targeted gene panels, very often in the absence of matching cancer-free genomic data. Here we examined the feasibility and accuracy of computational inference of genetic ancestry relying exclusively on cancer-derived data. A data synthesis framework was developed to optimize and assess the performance of the ancestry inference for any given input cancer-derived molecular profile. In its core procedure, the ancestral background of the profiled patient is replaced with one of any number of individuals with known ancestry. The data synthesis framework is applicable to multiple profiling platforms, making it possible to assess the performance of inference specifically for a given molecular profile and separately for each continental-level ancestry; this ability extends to all ancestries, including those without statistically sufficient representation in the existing cancer data. The inference procedure was demonstrated to be accurate and robust in a wide range of sequencing depths. Testing of the approach in four representative cancer types and across three molecular profiling modalities showed that continental-level ancestry of patients can be inferred with high accuracy, as quantified by its agreement with the gold standard of deriving ancestry from matching cancer-free molecular data. This study demonstrates that vast amounts of existing cancer-derived molecular data are potentially amenable to ancestry-oriented studies of the disease without requiring matching cancer-free genomes or patient self-reported ancestry.

SIGNIFICANCE

The development of a computational approach that enables accurate and robust ancestry inference from cancer-derived molecular profiles without matching cancer-free data provides a valuable methodology for genetic ancestry-oriented cancer research.

A cervical cancer biorepository for pharmacogenomics research in Zimbabwe

BACKGROUND

Research infrastructures such as biorepositories are essential to facilitate genomics and its growing applications in health research and translational medicine in Africa. Using a cervical cancer cohort, this study describes the establishment of a biorepository consisting of biospecimens and matched phenotype data for use in genomic association analysis and pharmacogenomics research.

METHOD

Women aged > 18 years with a recent histologically confirmed cervical cancer diagnosis were recruited. A workflow pipeline was developed to collect, store, and analyse biospecimens comprising donor recruitment and informed consent, followed by data and biospecimen collection, nucleic acid extraction, storage of genomic DNA, genetic characterization, data integration, data analysis and data interpretation. The biospecimen and data storage infrastructure included shared -20 °C to -80 °C freezers, lockable cupboards, secured access-controlled laptop, password protected online data storage on OneDrive software. The biospecimen or data storage, transfer and sharing were compliant with the local and international biospecimen and data protection laws and policies, to ensure donor privacy, trust, and benefits for the wider community.

RESULTS

This initial establishment of the biorepository recruited 410 women with cervical cancer. The mean (± SD) age of the donors was 52 (± 12) years, comprising stage I (15%), stage II (44%), stage III (47%) and stage IV (6%) disease. The biorepository includes whole blood and corresponding genomic DNA from 311 (75.9%) donors, and tumour biospecimens and corresponding tumour DNA from 258 (62.9%) donors. Datasets included information on sociodemographic characteristics, lifestyle, family history, clinical information, and HPV genotype. Treatment response was followed up for 12 months, namely, treatment-induced toxicities, survival vs. mortality, and disease status, that is disease-free survival, progression or relapse, 12 months after therapy commencement.

CONCLUSION

The current work highlights a framework for developing a cancer genomics cohort-based biorepository on a limited budget. Such a resource plays a central role in advancing genomics research towards the implementation of personalised management of cancer.

Analysis of germline-driven ancestry-associated gene expression in cancers

Differential mRNA expression between ancestry groups can be explained by both genetic and environmental factors. We outline a computational workflow to determine the extent to which germline genetic variation explains cancer-specific molecular differences across ancestry groups. Using multi-omics datasets from The Cancer Genome Atlas (TCGA), we enumerate ancestry-informative markers colocalized with cancer-type-specific expression quantitative trait loci (e-QTLs) at ancestry-associated genes. This approach is generalizable to other settings with paired germline genotyping and mRNA expression data for a multi-ethnic cohort.

For complete details on the use and execution of this protocol, please refer to Carrot-Zhang et al. (2020), Robertson et al. (2021), and Sayaman et al. (2021).

•

Protocol for obtaining controlled access TCGA datasets

•

Protocols for quality control analysis and genotype imputation of TCGA germline data

•

Statistical analysis for determining ancestry-associated SNPs

•

Determination of ancestry-associated germline genetic variation driving mRNA expression

Publisher’s note: Undertaking any experimental protocol requires adherence to local institutional guidelines for laboratory safety and ethics.

SL-Cloud: A Cloud-based resource to support synthetic lethal interaction discovery

Synthetic lethal interactions (SLIs), genetic interactions in which the simultaneous inactivation of two genes leads to a lethal phenotype, are promising targets for therapeutic intervention in cancer, as exemplified by the recent success of PARP inhibitors in treating BRCA1/2-deficient tumors. We present SL-Cloud, a new component of the Institute for Systems Biology Cancer Gateway in the Cloud (ISB-CGC), that provides an integrated framework of cloud-hosted data resources and curated workflows to enable facile prediction of SLIs. This resource addresses two main challenges related to SLI inference: the need to wrangle and preprocess large multi-omic datasets and the availability of multiple comparable prediction approaches. SL-Cloud enables customizable computational inference of SLIs and testing of prediction approaches across multiple datasets. We anticipate that cancer researchers will find utility in this tool for discovery of SLIs to support further investigation into potential drug targets for anticancer therapies.

SL-Cloud: A Cloud-based resource to support synthetic lethal interaction discovery

Synthetic lethal interactions (SLIs), genetic interactions in which the simultaneous inactivation of two genes leads to a lethal phenotype, are promising targets for therapeutic intervention in cancer, as exemplified by the recent success of PARP inhibitors in treating BRCA1/2-deficient tumors. We present SL-Cloud, a new component of the Institute for Systems Biology Cancer Gateway in the Cloud (ISB-CGC), that provides an integrated framework of cloud-hosted data resources and curated workflows to enable facile prediction of SLIs. This resource addresses two main challenges related to SLI inference: the need to wrangle and preprocess large multi-omic datasets and the availability of multiple comparable prediction approaches. SL-Cloud enables customizable computational inference of SLIs and testing of prediction approaches across multiple datasets. We anticipate that cancer researchers will find utility in this tool for discovery of SLIs to support further investigation into potential drug targets for anticancer therapies.

Anatomic Neuroimaging Characteristics of Posterior Fossa Type A Ependymoma Subgroups

BACKGROUND AND PURPOSE

Posterior fossa type A (PFA) ependymomas have 2 molecular subgroups (PFA-1 and PFA-2) and 9 subtypes. Gene expression profiling suggests that PFA-1 and PFA-2 tumors have distinct developmental origins at different rostrocaudal levels of the brainstem. We, therefore, tested the hypothesis that PFA-1 and PFA-2 ependymomas have different anatomic MR imaging characteristics at presentation.

MATERIALS AND METHODS

Two neuroradiologists reviewed the preoperative MR imaging examinations of 122 patients with PFA ependymomas and identified several anatomic characteristics, including extension through the fourth ventricular foramina and encasement of major arteries and tumor type (midfloor, roof, or lateral). Deoxyribonucleic acid methylation profiling assigned ependymomas to PFA-1 or PFA-2. Information on PFA subtype from an earlier study was also available for a subset of tumors. Associations between imaging variables and subgroup or subtype were evaluated.

RESULTS

No anatomic imaging variable was significantly associated with the PFA subgroup, but 5 PFA-2c subtype ependymomas in the cohort had a more circumscribed appearance and showed less tendency to extend through the fourth ventricular foramina or encase blood vessels, compared with other PFA subtypes.

CONCLUSIONS

PFA-1 and PFA-2 ependymomas did not have different anatomic MR imaging characteristics, and these results do not support the hypothesis that they have distinct anatomic origins. PFA-2c ependymomas appear to have a more anatomically circumscribed MR imaging appearance than the other PFA subtypes; however, this needs to be confirmed in a larger study.

Integrative modeling identifies genetic ancestry-associated molecular correlates in human cancer

Cellular and molecular aberrations contribute to the disparity of human cancer incidence and etiology between ancestry groups. Multiomics profiling in The Cancer Genome Atlas (TCGA) allows for querying of the molecular underpinnings of ancestry-specific discrepancies in human cancer. Here, we provide a protocol for integrative associative analysis of ancestry with molecular correlates, including somatic mutations, DNA methylation, mRNA transcription, miRNA transcription, and pathway activity, using TCGA data. This protocol can be generalized to analyze other cancer cohorts and human diseases.

For complete details on the use and execution of this protocol, please refer to Carrot-Zhang et al. (2020).

•

Protocols for ancestry associations with TCGA molecular data

•

Protocols for ancestry associations with oncogenic pathways

•

Statistical and power analysis for determining significant associations

•

Key considerations of potential confounding factors

Comprehensive Analysis of Genetic Ancestry and Its Molecular Correlates in Cancer

We evaluated ancestry effects on mutation rates, DNA methylation, and mRNA and miRNA expression among 10,678 patients across 33 cancer types from The Cancer Genome Atlas. We demonstrated that cancer subtypes and ancestry-related technical artifacts are important confounders that have been insufficiently accounted for. Once accounted for, ancestry-associated differences spanned all molecular features and hundreds of genes. Biologically significant differences were usually tissue specific but not specific to cancer. However, admixture and pathway analyses suggested some of these differences are causally related to cancer. Specific findings included increased FBXW7 mutations in patients of African origin, decreased VHL and PBRM1 mutations in renal cancer patients of African origin, and decreased immune activity in bladder cancer patients of East Asian origin.

COMP-04. MODELING PRECISION ONCOLOGY FOR GLIOBLASTOMA THROUGH INTEGRATION OF DESCRIPTIVE, FUNCTIONAL, AND NETWORK-BASED GENOMICS

Precision oncology is largely based on the notion that identification and targeting of oncogenic drivers will lead to improved clinical outcomes. However, the promise of precision oncology awaits to be fulfilled for many cancers, including Glioblastoma (GBM), where identification of oncogenic drivers has yet to improve survival rates. Here, we have attempted to systematically identify GBM vulnerabilities by performing genome-wide CRISRP-Cas9 lethality screens in patient-derived GBM stem-like cells (GSCs). In validation studies, we comprehensively retested GSC-specific hits in multiple GSC isolates, which were also genomically profiled (e.g. RNA-seq, exome-seq, CNV), and further integrated these data with CRISPR-Cas9 lethality screens from over 500 human cell lines from the Broad Institute’s CRISPR Avana dataset. As a result, we have begun making GBM dependency predictions and functional associations for top scoring hits, including: tumor developmental subtype; loss of functional redundancy with other genes/proteins; cancer-specific subnetworks of genes involved in mitochondrial protein turnover and membrane trafficking; and genes of unknown function essential for subset of GBMs. A few examples of these categories include the following scenarios. We find ADAR (Adenosine Deaminase RNA Specific) gene dependency is associated with the mesenchymal GBM subtype. The EFR3Agene, which has roles in maintaining active pools of phosphatidylinositol 4-kinase, appears required when the expression of its paralog EFR3Bis low or absent in tumor cells. The F-box protein-encoding gene FBXO42appears non-essential to most human cells lines and neural stem cells, but when knocked out in sensitive GSCs causes mitotic arrest, mitotic catastrophe, and cell death. While still a work in progress, we hope to use these results as a foundation for exploring and illuminating patient-specific molecular vulnerabilities for brain tumors. The results also underscore the need for integration of functional genetic approaches, where gene activities are inhibited, into precision oncology paradigms.

Genomic and Molecular Landscape of DNA Damage Repair Deficiency across The Cancer Genome Atlas

DNA damage repair (DDR) pathways modulate cancer risk, progression, and therapeutic response. We systematically analyzed somatic alterations to provide a comprehensive view of DDR deficiency across 33 cancer types. Mutations with accompanying loss of heterozygosity were observed in over 1/3 of DDR genes, including TP53 and BRCA1/2. Other prevalent alterations included epigenetic silencing of the direct repair genes EXO5, MGMT, and ALKBH3 in ∼20% of samples. Homologous recombination deficiency (HRD) was present at varying frequency in many cancer types, most notably ovarian cancer. However, in contrast to ovarian cancer, HRD was associated with worse outcomes in several other cancers. Protein structure-based analyses allowed us to predict functional consequences of rare, recurrent DDR mutations. A new machine-learning-based classifier developed from gene expression data allowed us to identify alterations that phenocopy deleterious TP53 mutations. These frequent DDR gene alterations in many human cancers have functional consequences that may determine cancer progression and guide therapy.

DNA Methylation Dynamics of Germinal Center B Cells Are Mediated by AID

Changes in DNA methylation are required for the formation of germinal centers (GCs), but the mechanisms of such changes are poorly understood. Activation-induced cytidine deaminase (AID) has been recently implicated in DNA demethylation through its deaminase activity coupled with DNA repair. We investigated the epigenetic function of AID in vivo in germinal center B cells (GCBs) isolated from wild-type (WT) and AID-deficient (Aicda(-/-)) mice. We determined that the transit of B cells through the GC is associated with marked locus-specific loss of methylation and increased methylation diversity, both of which are lost in Aicda(-/-) animals. Differentially methylated cytosines (DMCs) between GCBs and naive B cells (NBs) are enriched in genes that are targeted for somatic hypermutation (SHM) by AID, and these genes form networks required for B cell development and proliferation. Finally, we observed significant conservation of AID-dependent epigenetic reprogramming between mouse and human B cells.

IL10 receptor is a novel therapeutic target in DLBCLs

Diffuse large B-cell lymphoma (DLBCL) is a biologically and clinically heterogeneous disease with marked genomic instability and variable response to conventional R-CHOP (rituximab, cyclophosphamide, doxorubicin, vincristine and prednisone) chemotherapy. More clinically aggressive cases of DLBCLs have high level of circulating interleukin 10 (IL10) cytokine and evidence of activated intracellular STAT3 (signal transducer and activator of transcription 3) signaling. We investigated the role of IL10 and its surface receptor in supporting the neoplastic phenotype of DLBCLs. We determined that IL10RA gene is amplified in 21% and IL10RB gene in 10% of primary DLBCLs. Gene expression of IL10, IL10RA and IL10RB was markedly elevated in DLBCLs. We hypothesized that DLBCLs depend for their proliferation and survival on IL10-STAT3 signaling and that blocking the IL10 receptor (IL10R) would induce cell death. We used anti-IL10R blocking antibody, which resulted in a dose-dependent cell death in all tested activated B-cell-like subtype of DLBCL cell lines and primary DLBCLs. Response of germinal center B-cell-like subtype of DLBCL cell lines to anti-IL10R antibody varied from sensitive to resistant. Cells underwent cell cycle arrest, followed by induction of apoptosis. Cell death depended on inhibition of STAT3 and, to a lesser extent, STAT1 signaling. Anti-IL10R treatment resulted in interruption of IL10-IL10R autostimulatory loop. We thus propose that IL10R is a novel therapeutic target in DLBCLs.

Abstract LB-235: IL10 autoregulatory loop in DLBCLs: New biomarker and a therapeutic target

Diffuse Large B cell Lymphoma (DLBCL) is a common aggressive lymphoma that represents 30-40% of newly diagnosed cases of non-Hodgkin Lymphomas, but accounts for up to 80% of lymphoma-related mortality. It is biologically and clinically heterogeneous disease with variable response to conventional R-CHOP chemotherapy. R-CHOP remains the standard first line therapy after decades of investigation, but is associated with frequent lack of response. It has been reported that more clinically aggressive cases of DLBCLs have constitutive activation of NF-kB and STAT3 and have higher level of circulating IL10 cytokine in patient's peripheral. We further investigated the role of IL10 and its surface receptor in supporting the neoplastic phenotype of DLBCL. We measured and analyzed copy number changes using SNP array on a subset of 91 primary DLBCLs and identified broad regions of genomic amplification and deletion in this cohort using the GISTIC algorithm and determined that Il10RA is amplified in 17% and IL10RB in 8 % of primary DLBCLs. Gene expression for all 3 genes is markedly elevated, as determined using Affymetrix

HG U133 plus 2.0 array data on 59 primary DLBCLs: up to 3 fold for IL10RA, and more than 10 fold for IL10RB and IL10 cytokine as compared to normal Germinal center B cells (NGCB)(all t-test, p<0.01). We also confirmed that IL10R protein expression is elevated using Tissue microarray (TMA) with primary DLBCLs (independent cohort, n=80). Immunohostochemistry on TMA revealed that up to 73% of DLBCLs have overexpression of IL10R several fold above the expression level in normal B cells. ABC DLBCLs tend to have higher level of expression for all 3 genes as compared to GCB DLBCLs.

We thus hypothesized that DLBCLs are dependent on the feed-forward autostimulatory loop that starts from autocrine IL10 stimulation through overexpressed receptor leading to cell proliferation and that bliocking the receptor will lead to cell death.

We tested the effect of blocking the receptor using anti-IL10R Ab in a panel of 12 cell lines and 5 primary DLBCLs cultured ex-vivo. The Ab effect was dose-dependent and cell death ensued after 1-3 days of treatment. Within 3 days of treatment with 1 ug/ml most cell lines had reduced viability by more than 50%, and after treatment with 10 ug/ml all cell lines were more than 90% dead. On day 3 massive induction of apoptosis was detected in all DLBCLs using standard approaches: measuring Annexin V/ DAPI by flow cytometry and PARP-1 cleavage by western blot.

We determined that blocking IL10R results in specific inhibition of signaling through JAK1/2 and loss of phosphorylation at STAT3Y705 immediately after treatment and inhibition of signaling through MAPK and phosphorylation of STAT3S727 at later treatment time points. The inhibition of signaling is sustained for days with only one drug treatment leading to induction of apoptosis. We observed downregulation of the known transcriptional targets of STAT3 that are crucial for maintaining cell cycle and proliferation like CCND1, CCND2, CMYC, JUNB. Anti-IL10R treatment resulted in significant downregulation of IL10 and IL10RA transcription, thus leading to interruption of IL10-IL10R autostimulatory loop. We thus propose that IL10R is a novel therapeutic target in DLBCLs that allows easy detection and targeting. Our findings warranty further animal studies and development of humanized antibody for clinical use in patients.

Citation Format: Wendy Beguelin, Seema Sawh, Nyasha Chambwe, Huimin Geng, Yanwen Jiang, Pilar M. Dominguez, Wayne Tam, Rita Shaknovich. IL10 autoregulatory loop in DLBCLs: New biomarker and a therapeutic target. [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 LB-235. doi:10.1158/1538-7445.AM2014-LB-235

GobyWeb: simplified management and analysis of gene expression and DNA methylation sequencing data

We present GobyWeb, a web-based system that facilitates the management and analysis of high-throughput sequencing (HTS) projects. The software provides integrated support for a broad set of HTS analyses and offers a simple plugin extension mechanism. Analyses currently supported include quantification of gene expression for messenger and small RNA sequencing, estimation of DNA methylation (i.e., reduced bisulfite sequencing and whole genome methyl-seq), or the detection of pathogens in sequenced data. In contrast to previous analysis pipelines developed for analysis of HTS data, GobyWeb requires significantly less storage space, runs analyses efficiently on a parallel grid, scales gracefully to process tens or hundreds of multi-gigabyte samples, yet can be used effectively by researchers who are comfortable using a web browser. We conducted performance evaluations of the software and found it to either outperform or have similar performance to analysis programs developed for specialized analyses of HTS data. We found that most biologists who took a one-hour GobyWeb training session were readily able to analyze RNA-Seq data with state of the art analysis tools. GobyWeb can be obtained at http://gobyweb.campagnelab.org and is freely available for non-commercial use. GobyWeb plugins are distributed in source code and licensed under the open source LGPL3 license to facilitate code inspection, reuse and independent extensions http://github.com/CampagneLaboratory/gobyweb2-plugins.

Differential gene body methylation and reduced expression of cell adhesion and neurotransmitter receptor genes in adverse maternal environment

Early life adversity, including adverse gestational and postpartum maternal environment, is a contributing factor in the development of autism, attention deficit hyperactivity disorder (ADHD), anxiety and depression but little is known about the underlying molecular mechanism. In a model of gestational maternal adversity that leads to innate anxiety, increased stress reactivity and impaired vocal communication in the offspring, we asked if a specific DNA methylation signature is associated with the emergence of the behavioral phenotype. Genome-wide DNA methylation analyses identified 2.3% of CpGs as differentially methylated (that is, differentially methylated sites, DMSs) by the adverse environment in ventral-hippocampal granule cells, neurons that can be linked to the anxiety phenotype. DMSs were typically clustered and these clusters were preferentially located at gene bodies. Although CpGs are typically either highly methylated or unmethylated, DMSs had an intermediate (20-80%) methylation level that may contribute to their sensitivity to environmental adversity. The adverse maternal environment resulted in either hyper or hypomethylation at DMSs. Clusters of DMSs were enriched in genes that encode cell adhesion molecules and neurotransmitter receptors; some of which were also downregulated, indicating multiple functional deficits at the synapse in adversity. Pharmacological and genetic evidence links many of these genes to anxiety.

BDVal: reproducible large-scale predictive model development and validation in high-throughput datasets

High-throughput data can be used in conjunction with clinical information to develop predictive models. Automating the process of developing, evaluating and testing such predictive models on different datasets would minimize operator errors and facilitate the comparison of different modeling approaches on the same dataset. Complete automation would also yield unambiguous documentation of the process followed to develop each model. We present the BDVal suite of programs that fully automate the construction of predictive classification models from high-throughput data and generate detailed reports about the model construction process. We have used BDVal to construct models from microarray and proteomics data, as well as from DNA-methylation datasets. The programs are designed for scalability and support the construction of thousands of alternative models from a given dataset and prediction task.

AVAILABILITY AND IMPLEMENTATION

The BDVal programs are implemented in Java, provided under the GNU General Public License and freely available at http://bdval.campagnelab.org.