Search

The standard-of-care (SOC) for genomic testing of myeloid cancers primarily relies on karyotyping and fluorescent in situ hydridization (FISH) (cytogenetic analysis) and targeted gene panels (≤54 genes) that harbor hotspot pathogenic variants (molecular genetic analysis). Both cytogenetic and molecular testing workup is necessary for the identification and detection of large structural variants (SVs) and small variants like single nucleotide variants (SNV) and indels, respectively. Despite this combinatorial approach, ∼50% of myeloid cancer genomes remain cytogenetically normal, and the limited sequencing variant profiles obtained from targeted panels are unable to resolve the genetic etiology of these myeloid tumors. In this study, we evaluated the performance and clinical utility of optical genome mapping (OGM) and a 523-gene next-generation sequencing (NGS) panel for comprehensive genomic profiling of 15 myeloid tumors and compared it to SOC cytogenetic methods (karyotyping and FISH) and a 54-gene NGS panel. OGM and the 523-gene NGS panel were found to have an analytical concordance of 100% with karyotyping, FISH, and the 54-gene panel, respectively. Additionally, OGM better characterized and resolved the structural variants previously reported by karyotyping in five cases, such as identifying the genomic content of marker and ring chromosomes. OGM also identified several additional translocations and eleven copy number variations (CNVs), of which the CNVs were validated/confirmed by the 523-gene panel. The 523-gene panel identified seven additional clinically relevant SNVs (two tier 1A variants and five tier 2C variants, as per the ACMG/AMP guidelines) in four cases. The simultaneous visualization of SVs and small NGS detected sequence variants (SNVs and small indels) from OGM and 523-gene NGS panel, respectively in the NxClinical software v6.1 identified two clinically relevant compound heterozygous events in two samples. This study demonstrates the higher sensitivity, resolution, accuracy, and ability to reveal cryptic and clinically relevant novel variants in myeloid cancers as compared to SOC methodologies. Our cost-effective approach of using OGM and a 523-gene NGS panel for comprehensive genomic profiling of myeloid cancers will not only increase the yield of actionable targets leading to improved clinical outcomes but also help resolve our ongoing conundrum of apparently genomically normal myeloid cancers by providing more answers.

IMPORTANCE: Alterations in the IKZF1 gene drive B-cell acute lymphoblastic leukemia (B-ALL) but are not routinely used to stratify patients by risk because of inconsistent associations with outcomes. We describe a novel deletion in 22q11.22 that was consistently associated with very poor outcomes in patients with B-ALL with IKZF1 alterations. OBJECTIVE: To determine whether focal deletions within the λ variable chain region in chromosome 22q11.22 were associated with patients with B-ALL with IKZF1 alterations with the highest risk of relapse and/or death. DESIGN, SETTING, AND PARTICIPANTS: This cohort study included 1310 primarily high-risk pediatric patients with B-ALL who were taken from 6 independent clinical cohorts, consisting of 3 multicenter cohorts (AALL0232 [2004-2011], P9906 [2000-2003], and patients with Down syndrome who were pooled from national and international studies) and 3 single-institution cohorts (University of Utah [Salt Lake City], Children’s Hospital of Philadelphia [Philadelphia, Pennsylvania], and St. Jude Children’s Hospital [Memphis, Tennessee]). Data analysis began in 2011 using patients from the older studies first, and data analysis concluded in 2021. EXPOSURES: Focal 22q11.22 deletions. MAIN OUTCOMES AND MEASURES: Event-free and overall survival was investigated. The hypothesis that 22q11.22 deletions stratified the prognostic effect of IKZF1 alterations was formulated while investigating nearby deletions in VPREB1 in 2 initial cohorts (n = 270). Four additional cohorts were then obtained to further study this association (n = 1040). RESULTS: This study of 1310 patients with B-ALL (717 male [56.1%] and 562 female patients [43.9%]) found that focal 22q11.22 deletions are frequent (518 of 1310 [39.5%]) in B-ALL and inconsistent with physiologic V(D)J recombination. A total of 299 of 1310 patients with B-ALL had IKZF1 alterations. Among patients with IKZF1 alterations, more than half shared concomitant focal 22q11.22 deletions (159 of 299 [53.0%]). Patients with combined IKZF1 alterations and 22q11.22 deletions had worse outcomes compared with patients with IKZF1 alterations and wild-type 22q11.22 alleles in every cohort examined (combined cohorts: 5-year event-free survival rates, 43.3% vs 68.5%; hazard ratio [HR], 2.18; 95% CI, 1.54-3.07; P

Genomic based classification of cancers has been performed using gene expression and methylation patterns (e.g. well-established gene expression profiles used to sub-classify breast cancer). However, there has been very limited progress on classifying cancers based on their global CNV and/or LOH patterns. In this study, we present a novel neural network algorithm based on Self-Organizing Maps (SOMs) that creates a 2-dimensional map of cancer based on global CNV and LOH patterns. We applied this algorithm to 636 brain tumor samples from the TCGA project as part of the Glioblastoma (GBM) and Low-Grade Glioma (LGG) TCGA data sets. The CNV and LOH profiles for the samples was generated and manually curated to adjust for sample ploidy and call fragmentation due to normal cell contamination. Our algorithm generated a 25-node map of brain tumors, where different nodes represent a unique genome-wide profile and neighboring nodes are more similar to each other than more distant nodes. The generated map included nodes representing already known and prognostically important profiles such as 1p/19q co-deletion and combination of trisomy 7, monosomy 10, and homozygous deletion of p16. This is a significant validation of the approach since these profiles were generated completely automatically. In addition to these, new profiles were generated that were further investigated for prognostic value. We then projected all 636 samples back onto the learned map and were able to detect clusters of samples that generally fell within one cancer type, but also discovered some nodes shared by GBM and LGG samples indicating similarity in the CNV profile between these cancer types. We generated Kaplan-Meier survival plot for samples mapped to each node and were able to show statistically significant differences in survival based on the clustering. This approach has the possibility to create a unique CNV based classification of brain tumors for both prognostic and therapeutic applications. Citation Format: Megan Roytman, Soheil Shams. Brain cancer map: A neural network-based clustering of brain cancer samples based on genome-wide CNV and LOH patterns [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 2171.

A novel machine learning system trained on The Cancer Genome Atlas (TCGA) data has been designed to help with classification, prognostic measures, and monitoring treatment responses in cancer management. The utility of this system will be demonstrated using a set of retrospective samples. Cytogenetic analysis has long been utilized in clinical cancer care with conventional cytogenetic methods such as chromosomal banding and FISH still standard protocols for hematologic malignancies. With the advent of microarrays, smaller and more complex variants have been discovered in cancer samples but identifying the potential impact of many alterations such as copy number variation (CNV) and loss of heterozygosity (LOH) has not been fully exploited. A typical cancer cytogenetic report can list many tens of variants with unclear clinical implications. To bridge this gap and assist with determining the clinical implications of such alterations, we have developed a novel machine learning system that has been trained using data from The Cancer Genome Atlas (TCGA). The TCGA data has been processed using the SNP-FASST2 segmentation algorithm followed by manual review for ploidy adjustment yielding high quality regions of CNV and LOH for over 6000 cases across 29 cancer types. Our novel neural network used this processed data to arrange samples based on their CNV and LOH profile across a two-dimensional map. The proximity of one sample to another on this map corresponds to its similarity in terms of CNV and LOH profile. To apply this system in a clinical setting, we have designed a case review system that aligns a new case with the learned cancer map to provide information on the similarity of the case with ones in TCGA. This enables rapid verification of tumor type or detection of a different origin for the sample than the one reported. In addition, the system can create a prognostic score using clinical survival data for the samples in TCGA. Here we will demonstrate the utility of the system in classifying new cases against the learned cancer map and estimating prognostic measures. Citation Format: Megan Roytman, Viren Wasnikar, Paul An, Raja Kashevan, Shalini Verma, Soheil Shams. Improving cancer cytogenetic case review with a new machine learning system [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2094.

ABSTRACT Aim: Neurofibromatosis type 2 (NF2)-associated vestibular schwannomas have variable size at presentation which presents a unique challenge in NF2 patient management. Therefore, we investigated the molecular signature characteristic of the differences in size for improved individualized precise therapy. Materials & methods: RNA expression analysis was performed on 15 small and 27 large NF2-associated vestibular schwannoma tumors using a microarray analyzing over 47,000 transcripts. Results: A signature of 11 genes was found to be correlated with NF2 tumor size. Conclusion: We have identified the genetic hallmark that differentiates large NF2-associated tumors from smaller tumors. This is the first time that these genes have been shown to be the hallmark for NF2 tumor size.

The range of commercially available array platforms and analysis software packages is expanding and their utility is improving, making reliable detection of copy-number variants (CNVs) relatively straightforward. Reliable interpretation of CNV data, however, is often difficult and requires expertise. With our knowledge of the human genome growing rapidly, applications for array testing continuously broadening, and the resolution of CNV detection increasing, this leads to great complexity in interpreting what can be daunting data. Correct CNV interpretation and optimal use of the genotype information provided by single-nucleotide polymorphism probes on an array depends largely on knowledge present in various resources. In addition to the availability of host laboratories' own datasets and national registries, there are several public databases and Internet resources with genotype and phenotype information that can be used for array data interpretation. With so many resources now available, it is important to know which are fit-for-purpose in a diagnostic setting. We summarize the characteristics of the most commonly used Internet databases and resources, and propose a general data interpretation strategy that can be used for comparative hybridization, comparative intensity, and genotype-based array data. Hum Mutat 33: 930-940, 2012. (C) 2012 Wiley Periodicals, Inc.

Approximately 10% of patients with chronic lymphocytic leukaemia (CLL) have a family history of the disease or a related lymphoproliferative disorder, yet the relationship of familial CLL to genomic abnormalities has not been characterized in detail. We therefore studied 75 CLL patients, half familial and half sporadic, using high-resolution array comparative genomic hybridization (CGH), in order to better define the relationship of genomic abnormalities to familial disease and other biological prognostic factors. Our results showed that the most common high-risk deletion in CLL, deletion 11q, was significantly associated with sporadic disease. Comparison of familial to sporadic disease additionally identified a copy number variant region near the centromere on 14q, proximal to IGH@, in which gains were associated both with familial CLL, and with mutated IGHV and homozygous deletion of 13q. Homozygous deletion of 13q was also found to be associated with mutated IGHV and low expression of ZAP-70, and a significantly longer time to first treatment compared to heterozygous deletion or lack of alteration. This study is the first high resolution effort to investigate and report somatic genetic differences between familial and sporadic CLL.

At the Vancouver Prostate Centre (VPC), a personalized medicine approach is the focus to the understanding and treatment of prostate and other cancers. As a part of this approach, the xenograft program is used to identify and validate mechanisms and drivers of progress and therapeutic targets. VPC frequently uses Next-Generation Sequencing (NGS) technologies for copy number (CN) analysis. Many CN from NGS algorithms require a matched normal reference for CN estimation, however, a matched normal may not be possible for many reasons including unavailability of patient blood sample and poor quality of the matched sample in question. The correct CN algorithm for NGS data is critical to overcome these obstacles. Here we will share our results from analysis of Whole Exome Sequencing (WES) of prostate cancer patient-derived xenograft (PC PDX) samples using Nexus Copy Number software with BioDiscovery's BAM MSR algorithm. A pooled reference can be created using unmatched normal samples; however, when normal samples are not available, tumor samples from within the batch and/or run can be used to create a reference. BAM MSR derives CN and B-Allele Frequency (BAF) from WES, whole-exome sequencing (WGS), and targeted panels using a pooled reference. Due to an absence of normal samples, randomly selected tumor samples were used to build a reference file, which was used for baseline CNV results. By using a recursive refinement method, which incorporated evaluation of B-allele frequency (BAF) patterns among the pooled samples, a subset of tumor samples was selected to create a refined pooled reference using BAM MSR. All tumor samples were then subjected to CN estimation, which included adjustment based on tumor ploidy. While creating a normal reference including PC PDX tumor samples may seem counter intuitive due to enrichment of cancer cells and replacement of stromal component with mouse origin (e.g. fibroblasts, endothelium and immune cells), and high early recurrent copy number signatures, the BAM MSR algorithm was able to get results comparable to previous aCGH results. This is of special importance because the lack of normal tissues is commonplace in cancer research. Citation Format: Shawn Anderson, Zhiwei Che, Raja Keshavan, Andrea O'Hara, Dong Lin, Yuzhuo Wang, Colin Collins, Soheil Shams. Copy number estimation from whole-exome sequencing in tumors [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 3410.

Meningiomas are one of the most common adult brain tumors. For most patients, surgical excision is curative. However, up to 20% recur. Currently, the molecular determinants predicting recurrence and malignant transformation are lacking. We performed retrospective global genetic and genomic analysis of 85 meningioma samples of various grades. Copy number alterations were assessed by 100K SNP arrays and correlated with gene expression, proliferation indices, and clinical outcome. In addition to chromosome 22q loss, which was detected in the majority of clinical samples, chromosome 6q and 14q loss was significantly more common in recurrent tumors and was associated with anaplastic histology. Five "classes" of meningiomas were detected by gene expression analysis that correlated with copy number alterations, recurrent status, and malignant histology. These classes more accurately identified recurrent tumors relative to Ki-67 index and extent of surgical resection, and highlight substantial expression heterogeneity between meningiomas. These data offer the most complete description of the genomic landscape of meningiomas and provide broad genomic information that may be used to further stratify meningioma patients into prognostic risk groups.

Image processing is an important stage of every microarray experiment. Reliability of this stage strongly influences the results of data analysis performed on extracted gene expressions. Multiple methods related to array recognition, spot segmentation and measurement extraction have emerged in this area over past several years. Currently there are various commercial and freeware packages available, which perform microarray image analysis. This paper attempts to review microarray image analysis as a whole and to make some experimental comparison of several computational schemes for signal segmentation and measurement extraction. Also we provide a detailed discussion of automated image quality control for use with microarray images.

Next-generation sequencing (NGS) is mainly used to obtain sequence variants (SNVs). However, obtaining copy number results from NGS has gained momentum in both research and clinical applications. Targeted panel sequencing has been a popular method to achieve high depth of coverage for certain regions of interest at an affordable cost compared to whole genome sequencing. Shallow whole genome sequencing, where average read-depth can be as low as 0.1x, provides a cost savings-approach for identification of large copy number variant (CNV) events; it has been utilized in various application areas, including oncology. Here we introduce the BAM (MultiScale Reference) algorithm, currently in Nexus Copy Number, to function with shallow and targeted sequencing data, as well as WGS and WES, using a novel dynamic binning approach. This approach uses a Hidden Markov Model to segment the genome into target areas using the reads in targeted regions and the backbone areas using the off-target reads and additional areas. It uses coarse binning in the backbone areas that provides copy number base line as well as large copy number events and uses fine binning in target areas to provide high resolution copy number detection in targeted regions. Shallow WGS data and targeted panel NGS data, as well as WES with normal depth of coverage, were used for the testing. The results were compared with those from microarray and/or other algorithms in Nexus Copy Number, BAM ngCGH (matched) and BAM (pooled reference). GC correction schemes based on a range of window size and presence or absence of GC probe content were applied to the data and assessed for overall quality. Differences in overall read-depth resulted in variable sample quality across the cohorts, however most sample quality was adequate for copy number estimation and a quality threshold was assessed. Among the samples tested, the best quality after GC correction comes from the 50kb region size with or without the probes. Next, the copy number profiles of the samples from WES and microarray were compared for accuracy. Using microarray results as a reference for assessing calls greater than 5 MB, no false positive and one false negative call were observed; the single false negative call was attributable to low-level mosaicism in the tumor sample. Results indicate that relative copy number can be estimated and is comparable to the results achieved with microarray for the same targeted regions. This analysis series was then repeated using a secondary cohort of unrelated samples subjected to microarray and targeted panel NGS to validate results. The BAM (MultiScale Reference) method has been tested in a variety of cancer samples. This is an ideal tool for copy number estimation with NGS results in cancer samples because it provides a way for non-matched-pair analysis with genome, exome and targeted NGS. Citation Format: Andrea J. OHara, Zhiwei Che, Soheil Shams. Copy number estimation from targeted and shallow sequencing in cancer samples [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 3582. doi:10.1158/1538-7445.AM2017-3582

INTRODUCTION The goal of microarray experiments is quantifying the relative abundance of mRNA species contained within two or more biological samples sharing homologies with cDNA or oligonucleotides spotted (printed) onto a solid surface such as glass slides or nylon membranes. Automating the process has become a critical issue due to the magnitude of different cDNAs or oligonucleotides that can now be positioned on a single array. A single slide may contain up to 10,000 different sequences. Automation becomes even more critical considering that contemporary experiments often involve multi-slide protocols containing replicate samples and multiple controls. The current state-of-the-art utilizes an approach where each mRNA sample is reverse transcribed in the presence of a different fluorochrome-labeled dNTP and the resulting cDNAs are applied to the microarray slide. The relative binding of the labeled cDNAs is quantified by multiple scans of the slide with a laser set at the appropriate excitation wavelengths and recording an image of the slide in the emitted wavelengths. The scans are typically reported out as images in 16-bit grayscale TIFF file format. With this approach, the intensity of emitted light from an area of the slide containing spotted cDNA or oligonucleotide will be proportional to the amount of bound, fluorochrome-labeled probe over the dynamic range of the system. Processing of the scanned images of microarray slides consists of four steps: 1) measuring the signal intensity of each of the arrayed spots, 2) assessing the reliability of the data, 3) identifying any signal anomalies which may indicate problems in array fabrication or the performance of the hybridization steps, and 4) quantifying relative transcript abundance based on these intensities. A straightforward software solution for automating the first step would be superimposing a second, virtual image consisting of a grid of circles with both the same geometry as the spacing of the spots on the slide and the same diameter as the spots over the original image, followed by quantifying the pixels falling within and outside the individual circles. The former data would be the spot signals and the latter, background. The software would then package the data in a digital file format appropriate for direct import into other software packages for further analysis and databasing. Ideally, this last step would include associating the intensity measurements with unique names identifying each of the spotted cDNA sequences. The need for the second and third step listed above stems from complications, which occur during slide fabrication and the hybridization process. These complications require more robust automation software with the capability to provide data relating to image and spot quality. In terms of potential issues introduced during array fabrication, spot position is often inconsistent due to mechanical constraints in the spotting process. Mechanical constraints may also introduce spots with irregular shapes. Additionally, some drying rates may result in uneven distribution of the spotted sample leading to irregularities on the spot surface producing specular reflections in the image. It is also the case that extraneous signals may arise from splashes and drips of the DNA solutions occurring during printing as well as physical imperfections in glass substrate. Adding to the problem are environmental artifacts such as dust which may be introduced at any time during fabrication or hybridization and typically appear as very bright images in scanned arrays. These artificially high signals may occur in areas of the slide distinctly separated from arrayed spots as well as directly over true spots. Finally, artifacts in the images can be introduced during the hybridization procedure. The occurrence of spots with signal intensities below that of background is often ascribed to flaws in the hybridization protocol. Given these considerations, a fully functional software package for microarray image analysis would also need to automatically identify contamination and inconsistent spot placement and shape as well as flag spots of poor quality.

DNA array technology makes it possible to simultaneously study the expression of thousands of genes in a single experiment. DNA arrays are microscope slides (microarrays), or membrane filters (macroarrays) containing a large number of immobilized DNA samples. An array of cDNA-spots is subsequently probed with labeled cDNAs, which are obtained by reverse-transcriptase reaction from total RNA pools corresponding to the test and reference biological sources. Following the above hybridization step with dye-tagged or radioactively labeled probes, the DNA array is scanned to generate two images, each corresponding to one of the dye “colors.” A DNA array project typically requires iterations of series of processes, starting from experiment design and array fabrication, through array scanning, image analysis, and finally gene expression data analysis. In the maturation process of microarray technology, there are two kinds of challenge. One is to develop the hardware for conducting hybridization experiments. The other is to manage the massive amount of information associated with this technology, so that results can yield insight to the genomic functions in biological systems. The fundamental challenge from hardware has been mostly resolved. On the other hand, the informatics challenge has just started.

Managing vast amounts of information associated with DNA array technology presents a challenge. This article describes a synergistic analy- sis management (SAM) system, which integrates mi- croarray and laboratory data along with analysis steps to present a synergistic view to the researcher. We describe tools for data management in array fab- rication, automated image analysis, and array data mining. All the described modules allow for seamless flow of information and are connected through a da- tabase. SAM will enhance microarray projects at pharmaceutical and academic institutions, which face the problems of high throughput microarray data management. � 2000 John Wiley & Sons, Inc. Lab Robotics and Automation 12:317-327, 2000

P roducing microarray data starts with scanning in the glass, gel or plastic slides with a specialized scanner to obtain digital images of the results of an experiment after hybridization. With the help of image analysis software the DNA expression levels are then quantified. After the image processing and analysis step is completed we end up with a large number of quantified gene expression values. The data typically represents hundreds or thousands, in certain cases tens of thousands, of gene expressions across multiple experiments. To make sense of this much information it is unavoidable to use various visualization and statistical analysis techniques. One of the most typical microarray data analysis goals is to find statistically significant up or down regulated genes, in other words outliers or ‘interestingly’ behaving genes in the data. Other possible goals could be to find functional groupings of genes by discovering similarity or dissimilarity among gene expression profiles, or predicting the biochemical and physiological pathways of previously uncharacterized genes.

The Cancer Genome Atlas (TCGA) contains various types of genomic data from a wide variety of cancers, including several rare tumor types. Here we analyze copy number, sequence variant and RNA-Seq data from TCGA for kidney chromophobe (KICH). Level 1 (raw) data, rather than Level 3 (segmented) data, from TCGA was used for an integrated analysis of copy number and gene expression profiles of the two cancers. Using the copy number and SNP probes, re-processed tumor profiles were more consistent with a control set in terms of median number of copy number events, sample ploidy, and breakpoint genes than with the published level 3 TCGA data. Probe-level data were analyzed using Nexus Copy Number SNP-FASST2 algorithm (a multi-state HMM algorithm), with systematic correction applied to correct for GC biases. Additionally we performed manual baseline adjustment to correct for sample ploidy based on whole-genome B-allele frequency data for each sample. Overall, the median number of copy number events in the KICH TCGA data set was reduced from 408 (in the level 3 set) to 85. After manual inspection, more than 87% of the TCGA KICH samples available at level 3 were found to have incorrect baseline ploidy assignments. Given KICH samples are known to have low ploidy overall, this step was critical for downstream analysis. RNA-Seq results for the entire cohort were evaluated to obtain normalized relative RNA expression. Somatic sequence variation results from whole exome sequencing and relative RNA expression were integrated with the individual copy number profiles to yield integrated per-sample results across all three data modalities. Aggregate analysis indicates highly recurrent losses (50-85% of samples) on chromosomes 1, 2, 6, 10, 13, 17 and 21. Lower level recurrent losses (∼15-20% of samples) were identified on chromosomes 3, 5, 8, 9, 11 and 18. Concordance analysis revealed statistically significant subsets of samples with co-occurring lower level recurrent losses. Aggregate sequence mutation analysis of chromosomes among 3, 5, 8, 9, 11 and 18 did not identify any mutational hotspots. Integration with RNA-Seq expression data from each tumor type revealed statistically significant correlations (p Citation Format: Andrea J. O’Hara, Zhiwei Che, Soheil Shams. Integrated analysis of copy number, sequence variant and gene expression data in kidney chromophobe cohort. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 92.

Next-generation sequencing (NGS) is mainly used to obtain sequence variants (SNVs). However, obtaining copy number results from NGS has gained momentum in both research and clinical applications. While microarray has traditionally been the gold standard for copy number analysis, NGS is rapidly gaining momentum as the first line analysis for tumor samples. The genomic content captured may vary from batch to batch, where targeted regions may include the full genome, all exonic regions or may be limited to exonic regions in specific genes that have a clear diagnosis, are actionable with prescription drugs or compounds in clinical trials, or have known impacts on prognosis or outcome. Thus, obtaining copy number information from a range of NGS methods is needed. As a result, we have developed a method called BAM (pooled reference) which only requires the loading of BAM files and the NGS design file to generate copy number results. BAM (pooled reference) can derive copy number results from Whole Genome Sequencing (WGS), Whole Exome Sequencing (WES), and targeted panel NGS data. This method builds a reference file out of a pool of BAM files that are either from normal controls or from unrelated experimental samples, then generates copy number calls for each experimental sample. In combination with the VCF files that contain the sequence mutations, an integrated analysis of both events can be easily carried out. To test this algorithm, five colon adenocarcinoma whole exome sequencing samples were processed through the BAM (pooled reference) algorithm in Nexus Copy Number 8.0. GC correction schemes based on a range of window size and presence or absence of GC probe content were applied to the data and assessed for overall quality. Differences in overall read-depth resulted in variable sample quality across the cohorts, however most sample quality was adequate for copy number estimation and a quality threshold was assessed. Among the samples tested, the best quality after GC correction comes from the 50kb region size with or without the probes. Next, the copy number profiles of the TCGA COAD samples from WES and microarray were compared for accuracy. Using microarray results as a reference for assessing calls greater than 5 MB, no false positive and one false negative call were observed; the single false negative call was attributable to low-level mosaicism in the tumor sample. Results indicate that relative copy number can be estimated and is comparable to the results achieved with microarray for the same targeted regions. This analysis series was then repeated using a secondary cohort of unrelated samples subjected to microarray and targeted panel NGS to validate results. The BAM (pooled reference) method has been tested in a variety of cancer samples. This is an ideal tool for copy number estimation with NGS results in cancer samples because it provides a way for non-pair matched analysis with genome, exome and targeted NGS. Citation Format: Soheil Shams, Andrea O’Hara, Zhiwei Che. Copy number estimation of cancer samples with genome, exome and targeted panel next generation sequencing. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 3619.

High-throughput implementations of neural network models are required to transfer the technology from small prototype research problems into large-scale "real-world" applications. The flexibility of these implementations in accommodating for modifications to the neural network computation and structure is of paramount importance. The performance of many implementation methods today is greatly dependent on the density and the interconnection structure of the neural network model being implemented. A principal contribution of this paper is to demonstrate an implementation method which exploits maximum amount of parallelism from neural computation, without enforcing stringent conditions on the neural network interconnection structure, to achieve this high implementation efficiency. We propose a new reconfigurable parallel processing architecture, the Dynamically Reconfigurable Extended Array Multiprocessor (DREAM) machine, and an associated mapping method for implementing neural networks with regular interconnection structures. Details of the system execution rate calculation as a function of the neural network structure are presented. Several example neural network structures are used to demonstrate the efficiency of our mapping method and the DREAM machine architecture on implementing diverse interconnection structures. We show that due to the reconfigurable nature of the DREAM machine, most of the available parallelism of neural networks can be efficiently exploited. >

Fast synaptic plasticity, used to associate topologically ordered features in an input image to those of previously learned objects, has been previously proposed as a possible model for object recognition (von der Malsburg & Bienenstock, 1987, Europhysics Letters, 3(11), 1243–1249). In this paper, it is argued that in addition to rapid link dynamics, fast receptive field size dynamics are necessary to automatically escape from poor local matches and also allow for simultaneous recognition of multiple objects. Furthermore, a feature locking mechanism with a properly designed hysteresis property is needed to handle complex, cluttered, and dynamic scenes. The multiple elastic modules (MEM) model, described in this paper, utilizes newly developed dynamics that locate and recognize a previously learned object based on expected spatial arrangement of local features. The MEM model can be viewed as using a deformable template of an object to search the input scene. Unlike many of the current artificial neural network models, the proposed MEM model attempts to capture many of the functions available in the biological visual system by providing mechanisms for: multi-model feature integration, generation and maintenance of focus of attention, multiresolution hierarchical searching, and top-down expectation driven processing coupled with bottom-up feature activation processing. In addition, the MEM dynamics, unlike similar template matching approaches (Konen et al., 1994, Neural Networks, 7(6/7), 1019–1030; Yuille et al., 1992, International Journal of Computer Vision, 8(2), 99–111), does not converge to false objects when there are no sufficiently familiar objects in the scene. The performance of the MEM model in detection and recognition of objects through a number of computer simulations is demonstrated.

Neural network models have attracted much attention recently by demonstrating their potential at being an effective paradigm for implementing human-like intelligent processing. Neural network models, applied to “real-world” problems, demand high processing rates. Fortunately, neural network models contain several inherently parallel computing structures which can be utilized for high throughput implementations on parallel processing architectures. In this paper we describe the basic computational requirements and the various interconnection structures that are used by neural network models. A number of inherently parallel aspects of neural computing are described in detail along with a description of their specific demands on the supporting parallel processing architecture. The main obstacle in achieving efficient parallel implementations of neural networks is shown to be associated with the difficulty in efficiently supporting the complex and widely differing interconnection structures used by various neural network models. In this paper we survey several proposed implementation techniques organized based on a taxonomy of neural network interconnection structures.

Approximately 10% of patients with chronic lymphocytic leukaemia (CLL) have a family history of the disease or a related lymphoproliferative disorder, yet the relationship of familial CLL to genomic abnormalities has not been characterized in detail. We therefore studied 75 CLL patients, half familial and half sporadic, using high-resolution array comparative genomic hybridization (CGH), in order to better define the relationship of genomic abnormalities to familial disease and other biological prognostic factors. Our results showed that the most common high-risk deletion in CLL, deletion 11q, was significantly associated with sporadic disease. Comparison of familial to sporadic disease additionally identified a copy number variant region near the centromere on 14q, proximal to IGH@, in which gains were associated both with familial CLL, and with mutated IGHV and homozygous deletion of 13q. Homozygous deletion of 13q was also found to be associated with mutated IGHV and low expression of ZAP-70, and a significantly longer time to first treatment compared to heterozygous deletion or lack of alteration. This study is the first high resolution effort to investigate and report somatic genetic differences between familial and sporadic CLL.

The Cancer Genome Atlas (TCGA) contains various types of genomic data from a wide variety of cancers, several of which affect the same tissue site. Here we analyze copy number and RNA-Seq data from TCGA for lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC). Level 1 (raw) data, rather than Level 3 (segmented) data, from TCGA was used for an integrated analysis of copy number and gene expression profiles of the two cancers. Using the copy number and SNP- probes, re-processed tumor profiles were more consistent with a control set in terms of median number of copy number events, sample ploidy, and breakpoint genes than with the published “level 3” TCGA data. Probe-level data were analyzed using Nexus Copy Number SNP-FASST2 (a multi-state HMM algorithm that uses both SNP and copy number probes in making state assignments), with systematic correction applied to correct for GC biases. Additionally we performed manual baseline adjustment to correct for sample ploidy based on whole-genome B-alelle frequency data for each sample. Overall, the median number of copy number events in the LUAD TCGA data set was reduced from 371 (in the level 3 set) to 299, and from 681 (in the level 3 set) to 177 in the LUSC TCGA data set. After manual inspection, more than 38% of the TCGA LUAD samples and 50% of the TCGA LUSC samples available at level 3 were found to have incorrect baseline ploidy assignments. The resultant re-analyzed copy number data sets were used for an integrated analysis between the two tumor types. Comprehensive comparative analysis using Fisher's Exact Test revealed statistically significant differences (percent differential = 25%, p Citation Format: Andrea J. OHara, Raja Keshavan, Zhiwei Che, Soheil Shams. An integrated comparative analysis of TCGA lung adenocarcioma and lung squamous cell carcinoma copy number and RNA-Seq expression data. [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 2978. doi:10.1158/1538-7445.AM2015-2978

Motivation: A crucial step in microarray data analysis is the selection of subsets of interesting genes from the initial set of genes. In many cases, especially when comparing a specific condition to a reference, the genes of interest are those which are differentially expressed. Two common methods for gene selection are: (a) selection by fold difference (at least n fold variation) and (b) selection by altered ratio (at least n standard deviations away from the mean ratio). Results: The novel method proposed here is based on ANOVA and uses replicate spots to estimate an empirical distribution of the noise. The measured intensity range is divided in a number of intervals. A noise distribution is constructed for each such interval. Bootstrapping is used to map the desired confidence levels from the noise distribution corresponding to a given interval to the measured log ratios in that interval. If the method is applied on individual arrays having replicate spots, the method can calculate an overall width of the noise distribution which can be used as an indicator of the array quality. We compared this method with the fold change and unusual ratio method. We also discuss the relationship with an ANOVA model proposed by Churchill et al. In silico experiments were performed while controlling the degree of regulation as well as the amount of noise. Such experiments show the performance of the classical methods can be very unsatisfactory. We also compared the results of the 2-fold method with the results of the noise sampling method using pre and post immortalization cell lines derived from the MDAH041 fibroblasts hybridized on Affymetrix GeneChip arrays. The 2-fold method reported 198 genes as upregulated and 493 genes as downregulated. The noise sampling method reported 98 gene upregulated and 240 genes downregulated at the 99.99% confidence level. The methods agreed on 221 genes downregulated and 66 genes upregulated. Fourteen genes from the subset of genes reported by both methods were all confirmed by Q-RT-PCR. Alternative assays on various subsets of genes on which the two methods disagreed suggested that the noise sampling method is likely to provide fewer false positives. Contact: sod@cs.wayne.edu * To whom correspondence should be addressed. † The authors wish it to be known that, in their opinion, the first two authors should be regarded as joint First Authors.

High-throughput implementations of neural networks are needed in order to expand the use of this technology from small research problems into practical 'real-world' applications. Due to the wide range of possible neural network paradigms and the rapid evolution of these models, high degree of implementation flexibility if essential. The Dynamically Reconfigurable Extended Array Multiprocessor (DREAM) Machine has been specifically designed for implementation of neural networks. The architecture offers sufficient flexibility for use on a wide range of neural network applications. The authors describe the basic computational and communicational requirements of neural network models. A mapping method is proposed that can take advantage of the DREAM Machine's reconfigurable interconnection network to achieve efficient implementations for a diverse range of neural network structures. The effectiveness of the architecture and the mapping method is demonstrated through the use of several examples and shown to be superior to previous systolic implementation methods. >

Using the copy number and SNP- probes available in the TCGA level 1 data, reprocessed tumor profiles were more consistent with a control set in terms of median number of copy number events, sample ploidy, and breakpoint genes than the published “level 3” TCGA data. Probe-level data were analyzed using Nexus Copy Number SNP-FASST2 (a multi-state HMM algorithm that uses both SNP and copy number probes in making state assignments), with systematic correction applied to correct for GC biases. Additionally we performed manual baseline adjustment to correct for sample ploidy based on whole-genome B-alelle frequency data for each sample. The median number of copy number events in the TCGA data set was reduced from 377 (in the level 3 set) to 193, with a corresponding reduction in variance, and in line with our control samples (median CN changes of 173 and 142). After manual inspection, more than 15% of the TCGA BRCA samples available at level 3 were found to have incorrect baseline ploidy assignments. The resultant reanalyzed data set, along with associated clinical annotations for each sample, is available through Nexus DB repository to the general scientific community and should provide greater utility for further research. Citation Format: Louis J. Culot, Andrea O'Hara, Raja Keshavan, Zhiwei Che, Soheil Shams. Re-analysis of breast invasive carcinoma (BRCA) TCGA copy number data improves tumor profiles. [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 5175. doi:10.1158/1538-7445.AM2014-5175

The task of visual object recognition is often complicated by the fact that a single 3-D object can undergo a number of transformations which can substantially alter its projection onto a 2-D surface, such as the retina. Such transformations include translation of the object in the visual field, changes in the size of the object, its orientation in the 2-D plane and the viewing perspective. For a general pattern recognition system to detect and recognize an object after such transformations, it must be able to associate widely differing patterns with the same object label. In this paper, a novel self-organizing model, called the Multiple Elastic Modules (MEM), is presented which attempts to solve this problem by searching a multi-dimensional space, where each axis is defined by one of the transformations (e.g scale, translation, rotation, etc.). A particular object of a specific size, orientation and spatial location is mapped onto a single point in this space. Of course, distortions and minor variations in an object's image will expand this point to a small localized area in this multi-dimensional space. Such a powerful representation scheme comes at a cost of high computational demand due to the combinatorially large search space. The MEM approach to solving this problem efficiently partitions the solution space to search the most promising areas for the correct match. Simulation results are presented on detecting a stick-figure object under translation, distortion, scale, and rotation transformations in a cluttered background.

The Cancer Genome Atlas (TCGA) data is available via the TCGA portal and includes copy number data and whole exome sequencing data. Nexus Copy Number 7 was used to evaluate level-1 approved SNP array data from the colon adenocarcinoma (COAD) data set. All samples were subjected to match-paired analysis to evaluate somatic copy number variation. Samples were subjected to extensive pre-processing, including systematic G/C wave correction, ploidy correction, gender correction and quality control. Corresponding level-3 somatic mutation data was used for an integrated analysis of copy number and sequence variation with the COAD data set. Comprehensive copy number analysis identified frequent (>35%) copy number gains of chr7, chr8q, chr13 and chr20, and losses of chr8p, chr17p, and chr18. GISTIC analysis further identified statistically significant focal changes which included loss of FHIT on chr3. Genes KRAS and BRAF were among the most frequently mutated genes. Integrated analysis identified several regions of copy number change that significantly co-occurred with BRAF mutation. Copy number profiles varied significantly between hypermutable and non-hypermutable samples. Potential chromothriptic events were also observed within the COAD dataset. An approach that can integrate structural and sequence variation among samples will be presented. Citation Format: Andrea O'Hara, Raja Keshavan, Louis Culot, Soheil Shams. Integrated analysis of sequence variations and copy number in TCGA colon adenocarcinoma data. [abstract]. In: Proceedings of the Third AACR International Conference on Frontiers in Basic Cancer Research; Sep 18-22, 2013; National Harbor, MD. Philadelphia (PA): AACR; Cancer Res 2013;73(19 Suppl):Abstract nr C33.

Molecular cytogenetic alterations, specifically monosomy 3, are strongly correlated with metastases and death in uveal melanoma (UVM). Although FISH can be used for the identification of monosomy 3, a subset of UVM exhibit only partial loss of chromosome 3 potentially missed with chromosome enumeration probes. Moreover, often limited material available can compromise detection of additional alterations with potential clinical relevance. Microarray analysis is an alternative method for the analysis of such specimens affording whole genome interrogation. In the current study we performed microarray to detect genomic changes in DNA from FFPET and FZT UVM samples (clinical trial NCT00952939). Of the available cases, 23.5% yielded DNA of sufficient quantity and quality to obtain interpretable microarray results, representing 28 patients. All but one case demonstrated significant abnormalities. Gains of 8q, consistent with an apparent i(8)(q10) karyotype, were the most common finding, seen in 21 patients. Monosomy 3 was detected by microarray in 15 cases. A single case showed partial loss of 3 (3q11.2q25.31), the only clinically significant abnormality detected in that case. All cases lacking chromosome 3 abnormalities showed a copy gain of 6p sharing a small 28.18 Mb region of overlap at 6p25.2p21.33. Additional findings included TERT and NEDD9 amplifications, and CDKN2A/B and LUM gene deletions. SNP analysis revealed three cases with unique regions of copy neutral LOH involving 5p15.33q35.3, 15q11.2q21.1 and one case with whole chromosome LOH for chromosomes 3, 4 and 6. The latter was detected in a FZT specimen, while the paired FFPET sample lacked evidence of LOH and showed monosomy 3, trisomy 4 and whole arm gains and losses of 6p and 6q, respectively, suggesting the presence of tumor heterogeneity. Microarray analysis has identified new recurrent abnormalities associated with UVM with potential relevance to UVM biology, diagnosis and prognosis. Conflict of Interest: Roger Schultz is an employee of Perkin Elmer.

Using data from The Cancer Genome Atlas (TCGA) project and HapMap consortium, we examined the hypothesis that existing copy number polymorphisms might predispose individuals to cancer and if different common CNVs bias toward different cancer types. This analysis was performed using a subset of the TCGA data, obtained from blood derived “normal” samples of cancer patients. All the data was generated using Affymetrix SNP 6.0 arrays (Affymetrix, Inc.) and processed using Nexus Copy Number ver. 6.1 (BioDiscovery, Inc.). The project contained 451 Glioblastoma (GBM), 353 Colon cancer, and 245 Skin cancer samples in addition to 211 HapMap samples which we used as reference since these individuals were supposed to be healthy. Using Fisher's Exact test, we compared each different cancer types to the HapMap samples to identify regions of copy number change with FDR corrected p-value of less than 0.05 and difference in frequency between each population of at least 10%. Many of the regions found in this process encompass regions identified as common CNV polymorphisms as reported in the Database of Genomic Variance (DGV). For example, a polymorphic region on 8p23.1 containing a number of genes from the Beta defensin family involved in defense response to bacterium were seen significantly gained in germline samples obtained from cancer patients as compared to the HapMap samples (11% in HapMap, 42% in Colon Cancer, 28% in GBM, and 32% in Skin cancer). Additionally, some interesting regions were identified that were not seen as polymorphic in the general population, did not show up in the HapMap samples or reported in DGV, but had high prevalence in the blood derived normal samples from cancer patients. Using gene enrichment analysis on genes discovered in these regions we identified a number of known cancer specific pathways that were highly enriched (e.g. KEGG Glioma and Melanoma pathways being statistically significantly enriched). Looking at the effected genes in these pathways we identified small focal and recurrent events that often only spanned a small portion of a gene covering a few exons. This included genes such as PIK3CA where a focal gain covering exons 10-14 were observed in none of the HapMap samples but in 21% of the germline of patients with Colon cancer, 9% of GBM and 14% of Skin cancer patients. A similar focal and recurrent gain was observed in KIT covering exons 2-7 with no gains in the HapMap samples and 55% in Colon, 15% in GBM, and 50% in Skin cancer patients. Similar observations were also made for other well-known cancer genes such as RAF1 and EGFR but to a lesser extent. We are in a process of comparing matched tumor/normal cases to further investigate possible evolution of these changes from the germline to the surrounding tissue (also available from the TCGA database) to the primary tumor samples. Citation Format: Soheil Shams, Raja Keshavan, Joshua D. Schiffman. Analysis of copy number and LOH in germline samples from different tumor types in The Cancer Genome Atlas (TCGA) project. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 3171. doi:10.1158/1538-7445.AM2013-3171