Saurabh Agarwal

The impact of noise on the scaling of collectives: a theoretical approach

The performance of parallel applications running on large clusters is known to degrade due to the interference of kernel and daemon activities on individual nodes, often referred to as noise. In this paper, we focus on an important class of parallel applications, which repeatedly perform computation, followed by a collective operation such as a barrier. We model this theoretically and demonstrate, in a rigorous way, the effect of noise on the scalability of such applications. We study three natural and important classes of noise distributions: The exponential distribution, the heavy-tailed distribution, and the Bernoulli distribution. We show that the systems scale well in the presence of exponential noise, but the performance goes down drastically in the presence of heavy-tailed or Bernoulli noise.

A Mouse Model of X-linked Intellectual Disability Associated with Impaired Removal of Histone Methylation

Mutations in a number of chromatin modifiers are associated with human neurological disorders. KDM5C, a histone H3 lysine 4 di- and tri-methyl (H3K4me2/3)-specific demethylase, is frequently mutated in X-linked intellectual disability (XLID) patients. Here, we report that disruption of the mouse Kdm5c gene recapitulates adaptive and cognitive abnormalities observed in XLID, including impaired social behavior, memory deficits, and aggression. Kdm5c-knockout brains exhibit abnormal dendritic arborization, spine anomalies, and altered transcriptomes. In neurons, Kdm5c is recruited to promoters that harbor CpG islands decorated with high levels of H3K4me3, where it fine-tunes H3K4me3 levels. Kdm5c predominantly represses these genes, which include members of key pathways that regulate the development and function of neuronal circuitries. In summary, our mouse behavioral data strongly suggest that KDM5C mutations are causal to XLID. Furthermore, our findings suggest that loss of KDM5C function may impact gene expression in multiple regulatory pathways relevant to the clinical phenotypes.

An Integrated Approach to Identifying Cis-Regulatory Modules in the Human Genome

In eukaryotic genomes, it is challenging to accurately determine target sites of transcription factors (TFs) by only using sequence information. Previous efforts were made to tackle this task by considering the fact that TF binding sites tend to be more conserved than other functional sites and the binding sites of several TFs are often clustered. Recently, ChIP-chip and ChIP-sequencing experiments have been accumulated to identify TF binding sites as well as survey the chromatin modification patterns at the regulatory elements such as promoters and enhancers. We propose here a hidden Markov model (HMM) to incorporate sequence motif information, TF-DNA interaction data and chromatin modification patterns to precisely identify cis-regulatory modules (CRMs). We conducted ChIP-chip experiments on four TFs, CREB, E2F1, MAX, and YY1 in 1% of the human genome. We then trained a hidden Markov model (HMM) to identify the labels of the CRMs by incorporating the sequence motifs recognized by these TFs and the ChIP-chip ratio. Chromatin modification data was used to predict the functional sites and to further remove false positives. Cross-validation showed that our integrated HMM had a performance superior to other existing methods on predicting CRMs. Incorporating histone signature information successfully penalized false prediction and improved the whole performance. The dataset we used and the software are available at http://nash.ucsd.edu/CIS/.

Compression tolerant watermarking for image verification

Digital watermarking is seen as a viable solution to authentication of multimedia data and hence its security, especially in a networked environment. We present a new watermarking technique to add a code to digital images in the spatial domain. This technique is further shown to be robust under common compression schemes, including lossy compression schemes. Watermark embedding is done keeping in mind the limitations of the human visual system. We have used a procedure for error diffusion so as to minimize the chances of image tampering. For the verification of the image, the original uncorrupted image is not required. Experimental results show that the watermark is robust to JPEG compression.

Impact of Job Allocation Strategies on Communication-Driven Coscheduling in Clusters

In this paper, we investigate the impact of three job allocation strategies on the performance of four coscheduling algorithms (SB, DCS, PB and CC) in a 16-node Linux cluster. The job allocation factors include Multi Programming Level (MPL), job placement, and communication intensity. The experimental results show that the blocking based coscheduling schemes (SB and CC) have better tolerance to different job allocation techniques compared to the spin based schemes (DCS and PB), and the local scheduling. The results strengthen the case for using blocking based coscheduling schemes in a cluster.

Performance Comparison of Coscheduling Algorithms for Non-Dedicated Clusters Through a Generic Framework

In this paper, we address several key issues in designing coscheduling algorithms for clusters. First, we propose a generic framework for deploying coscheduling techniques by providing a reusable and dynamically loadable kernel module. Second, we implement several communication-driven coscheduling algorithms [dynamic coscheduling (DCS), spin block (SB) and periodic boost (PB)] on a 16- node Linux cluster using the above framework. Third, with exhaustive experimentation using mixed workloads, we observe that unlike PB, which provided the best performance on a Solaris platform, the SB scheme outperforms all other techniques on a Linux platform. Finally, we investigate the impact of several job placement strategies, multiprogramming level (MPL), communication intensity and CPU and I/O intensive jobs on the performance of these coscheduling schemes. The experimental results show that the blocking-based coscheduling scheme (SB) has better tolerance to system workload variation compared with the spin-based schemes (DCS and PB), and overall, the blocking-based coscheduling scheme seems a better choice for non-dedicated Linux clusters.

LSD1/KDM1A Maintains Genome-wide Homeostasis of Transcriptional Enhancers

Transcriptional enhancers enable exquisite spatiotemporal control of gene expression in metazoans. Enrichment of mono-methylation of histone H3 lysine 4 (H3K4me1) is a major chromatin signature that distinguishes enhancers from gene promoters. Lysine Specific Demethylase 1 (LSD1, aka KDM1A), an enzyme specific for demethylating H3K4me2/me1, has been shown to “decommission” stem cell enhancers during the differentiation of mouse embryonic stem cells (mESC). However, the roles of LSD1 in undifferentiated mESC remain obscure. Here, we show that LSD1 occupies a large fraction of enhancers (63%) that are primed with binding of transcription factors (TFs) and H3K4me1 in mESC. In contrast, LSD1 is largely absent at latent enhancers, which are not yet primed by TF binding. Unexpectedly, LSD1 levels at enhancers exhibited a clear positive correlation with its substrate, H3K4me2 and enhancer activity. These enhancers gain additional H3K4 methylation upon the loss of LSD1 in mESC. The aberrant increase in H3K4me at enhancers was accompanied with increases in enhancer H3K27 acetylation and expression of enhancer RNAs (eRNAs) and their target genes. In post-mitotic neurons, loss of LSD1 resulted in premature activation of enhancers and genes that are normally induced after neuronal activation. These results demonstrate that LSD1 is a versatile suppressor of primed enhancers, and is involved in homeostasis of enhancer activity.

Identification of a Peptide-like Compound with Antimicrobial and Trypsin Inhibitory Activity from Seeds of Bottle Gourd (Lagenaria siceraria)

A low molecular mass peptide like compound with antimicrobial and trypsin inhibitory activity was isolated from the seeds of Lagenaria siceraria. It was purified by ion-exchange and reverse-phase chromatography. The molecular weight of the compound was 678.9 Dalton as determined by MALDI-MS. The infra-red absorbance at 1639 cm−1, characteristic of an amide bond, by FTIR spectroscopic studies, and absorption at 214 nm on spectrophotometer indicates the peptidic nature of the compound. The compound exhibited antimicrobial activity when tested against Escherichia coli with minimum inhibitory concentration of 20 μM, and trypsin inhibitory activity inhibiting trypsin at a molar ratio of 1:2.

Patient Mutations of the Intellectual Disability Gene KDM5C Downregulate Netrin G2 and Suppress Neurite Growth in Neuro2a Cells

The X-linked lysine (K)-specific demethylase 5C (KDM5C) gene plays an important role in brain development and behavior. It encodes a histone demethylase that is involved in gene regulation in neuronal differentiation and morphogenesis. When mutated, it causes neuropsychiatric symptoms, such as intellectual disability, delayed language development, epilepsy, and impulsivity. To better understand how the patient mutations affect neuronal development, we expressed KDM5C mutants in Neuro2a cells, a mouse neuroblastoma cell line. Retinoic acid (RA)-induced neurite growth was suppressed by the mutation KDM5CY751C, KDM5CH514A, and KDM5CF642L, but not KDM5CD87G or KDM5CA388P. RNA-seq analysis indicated an upregulation of genes important for neuronal development, such as Ntng2, Enah, Gas1, Slit2, and Dscam, in response to the RA treatment in control Neuro2a cells transfected with GFP or wild-type KDM5C. In contrast, in cells transfected with KDM5CY751C, these genes were not upregulated by RA. Ntng2 was downregulated in cells with KDM5C mutations, concordant with the lower levels of H3K4 methylation at its promoter. Moreover, knocking down Ntng2 in control Neuro2a cells led to the phenotype of short neurites similar to that of cells with KDM5CY751C, whereas Ntng2 overexpression in the mutant cells rescued the morphological phenotype. These findings provide new insight into the pathogenesis of phenotypes associated with KDM5C mutations.