MultiBench: Multiscale Benchmarks for Multimodal Representation Learning.

Learning multimodal representations involves integrating information from multiple heterogeneous sources of data. It is a challenging yet crucial area with numerous real-world applications in multimedia, affective computing, robotics, finance, human-computer interaction, and healthcare. Unfortunately, multimodal research has seen limited resources to study (1) generalization across domains and modalities, (2) complexity during training and inference, and (3) robustness to noisy and missing modalities. In order to accelerate progress towards understudied modalities and tasks while ensuring real-world robustness, we release MultiBench, a systematic and unified large-scale benchmark for multimodal learning spanning 15 datasets, 10 modalities, 20 prediction tasks, and 6 research areas. MultiBench provides an automated end-to-end machine learning pipeline that simplifies and standardizes data loading, experimental setup, and model evaluation. To enable holistic evaluation, MultiBench offers a comprehensive methodology to assess (1) generalization, (2) time and space complexity, and (3) modality robustness. MultiBench introduces impactful challenges for future research, including scalability to large-scale multimodal datasets and robustness to realistic imperfections. To accompany this benchmark, we also provide a standardized implementation of 20 core approaches in multimodal learning spanning innovations in fusion paradigms, optimization objectives, and training approaches. Simply applying methods proposed in different research areas can improve the state-of-the-art performance on 9/15 datasets. Therefore, MultiBench presents a milestone in unifying disjoint efforts in multimodal machine learning research and paves the way towards a better understanding of the capabilities and limitations of multimodal models, all the while ensuring ease of use, accessibility, and reproducibility. MultiBench, our standardized implementations, and leaderboards are publicly available, will be regularly updated, and welcomes inputs from the community.

Suppression of tumor growth via IGFBP3 depletion as a potential treatment in glioma.

OBJECTIVE: Despite intensive medical treatment, patients with glioblastoma (grade IV glioma [GBM]) have a low 5-year survival rate of 5.5%. In this study, the authors tried to improve currently used therapies by identification of a therapeutic target, IGFBP3, for glioma treatment. METHODS: IGFBP3 RNA expression in 135 patients newly diagnosed with glioma was correlated with clinicopathological factors. Immunohistochemical analysis was performed to determine IGFBP3 protein expression in glioma specimens. The effect of IGFBP3 depletion on cell proliferation was examined using IGFBP3 knockdown glioma cells. Intracranial infusion of IGFBP3 siRNAs was performed to evaluate the effect of IGFBP3 depletion in mouse intracranial xenograft models. RESULTS: We demonstrated higher IGFBP3 expression in GBM than in tumor margin and grade II glioma. IGFBP3 expression was not only positively correlated with tumor grades but also associated with tumor histology and IDH1/2 mutation status. Additionally, higher IGFBP3 expression predicted shorter overall survival in glioma and GBM proneural subgroup patients. In vitro cell culture studies suggested IGFBP3 knockdown suppressed cell proliferation and induced cell cycle G2/M arrest as well as apoptosis in glioma cells. Also, accumulation of DNA double-strand breaks and γH2AX was observed in IGFBP3 knockdown cells. IGFBP3 knockdown delayed in vivo tumor growth in mouse subcutaneous xenograft models. Furthermore, convection-enhanced delivery of IGFBP3 siRNA to mouse brain suppressed intracranial tumor growth and prolonged survival of tumor-bearing mice. CONCLUSIONS: Our findings suggest IGFBP3 predicts poor outcome of glioma patients and is a potential therapeutic target for which depletion of its expression suppresses tumor growth through inducing apoptosis and accumulation of DNA damage in glioma cells.

The skin microbiome of wound scars and unaffected skin in patients with moderate to severe burns in the subacute phase.

Although studies on skin microbiome of acute and chronic wounds abound, evidence on newly built microbial communities of subacute wounds remains scant. To characterize the skin microbiome of recently healed (scarred) burn wounds in relation to unaffected skin surfaces, we collected weekly swabs from patients with moderate to severe burns in the 3rd postburn month for 4 weeks in 2015. We performed skin type (moist, dry, and oily)-matched comparisons within six burn patients (43 pairs of swabs) and with 13 skin-healthy, control patients (22 pairs of samples) using 16S ribosomal RNA gene sequencing results. Results of comparative microbiome analysis showed that, there were no substantial variations in the microbial abundance (all p > 0.05) or composition (all p > 0.01, adjusted for multiple comparisons) between samples obtained from wound scars and those from unaffected surfaces of burn patients. Nor did we find significant temporal dynamics in microbial richness or diversity in burn samples (all p ≥ 0.05). However, samples from burn patients harbored more Firmicutes (median: 25.6%, interquartile range [IQR]: 14.3%-52.8%) than those of control patients (14.9%, IQR: 6.7%-27.0%; p: 0.016), even after adjusting for host age, sex, and skin type-matching (p: 0.026). The number of observed bacterial operational taxonomic units at the genus level was reduced in burn patients (median: 62, IRQ: 32-85) as compared to control patients (median: 128, IQR: 112-136; age-, skin type-adjusted p < 0.01). Meanwhile, estimates of community diversity and evenness for surveyed body sites of burn patients were higher than those of control patients (all adjusted p ≤ 0.05). With a much-reduced bacterial burden and a relative overgrowth of Staphylococcus spp., the skin microbiota of burn patients remained dysbiotic in the subacute phase as compared to that of skin-normal patients.

The rs16906252:C>T SNP is not associated with increased overall survival or temozolomide response in a Han-Chinese glioma cohort.

The methylation status of O-6-methylguanine-DNA methyltransferase (MGMT) is associated with the prognosis in gliomas and in other cancers. Recent studies showed that rs16906252, an SNP in the MGMT promoter, is associated with promoter methylation and is a predictor of the overall survival time (OST) and the response to temozolomide (TMZ) treatment. However, these findings haven't been systematically investigated in the Han-Chinese population. We analyzed the relevance between rs16906252 polymorphisms, the MGMT methylation status, and the OST in 72 Han-Chinese gliomas patients. The MGMT promoter methylation was measured by bisulfite conversion followed by pyro-sequencing, while rs16906252 was measured by restriction endonuclease digestion. Contrary to the previous findings, we found no association between rs16906252 genotypes and promoter methylation on MGMT. The lower-grade glioma (LGGs) patients carrying the C allele with rs16906252 showed a surprisingly better OST (P = 0.04). Furthermore, the LGG patients carrying hypo-methylated MGMT promoter and rs16906252 T allele showed significantly poorer prognosis. The prognostic benefit of MGMT promoter methylation and genotypes on gliomas patients is marginal. A new molecular stratified patient grouping of LGGs is potentially associated with poorer OST. Active MGMT might have a protective role in LGG tumors, enabling evolution to severe malignancy.

JAKMIP1, a Novel Regulator of Neuronal Translation, Modulates Synaptic Function and Autistic-like Behaviors in Mouse.

Autism spectrum disorder (ASD) is a heritable, common neurodevelopmental disorder with diverse genetic causes. Several studies have implicated protein synthesis as one among several of its potential convergent mechanisms. We originally identified Janus kinase and microtubule-interacting protein 1 (JAKMIP1) as differentially expressed in patients with distinct syndromic forms of ASD, fragile X syndrome, and 15q duplication syndrome. Here, we provide multiple lines of evidence that JAKMIP1 is a component of polyribosomes and an RNP translational regulatory complex that includes fragile X mental retardation protein, DEAD box helicase 5, and the poly(A) binding protein cytoplasmic 1. JAKMIP1 loss dysregulates neuronal translation during synaptic development, affecting glutamatergic NMDAR signaling, and results in social deficits, stereotyped activity, abnormal postnatal vocalizations, and other autistic-like behaviors in the mouse. These findings define an important and novel role for JAKMIP1 in neural development and further highlight pathways regulating mRNA translation during synaptogenesis in the genesis of neurodevelopmental disorders.

Correspondence between Resting-State Activity and Brain Gene Expression.

The relationship between functional brain activity and gene expression has not been fully explored in the human brain. Here, we identify significant correlations between gene expression in the brain and functional activity by comparing fractional amplitude of low-frequency fluctuations (fALFF) from two independent human fMRI resting-state datasets to regional cortical gene expression from a newly generated RNA-seq dataset and two additional gene expression datasets to obtain robust and reproducible correlations. We find significantly more genes correlated with fALFF than expected by chance and identify specific genes correlated with the imaging signals in multiple expression datasets in the default mode network. Together, these data support a population-level relationship between regional steady-state brain gene expression and resting-state brain activity.

Mutation of senataxin alters disease-specific transcriptional networks in patients with ataxia with oculomotor apraxia type 2.

Senataxin, encoded by the SETX gene, contributes to multiple aspects of gene expression, including transcription and RNA processing. Mutations in SETX cause the recessive disorder ataxia with oculomotor apraxia type 2 (AOA2) and a dominant juvenile form of amyotrophic lateral sclerosis (ALS4). To assess the functional role of senataxin in disease, we examined differential gene expression in AOA2 patient fibroblasts, identifying a core set of genes showing altered expression by microarray and RNA-sequencing. To determine whether AOA2 and ALS4 mutations differentially affect gene expression, we overexpressed disease-specific SETX mutations in senataxin-haploinsufficient fibroblasts and observed changes in distinct sets of genes. This implicates mutation-specific alterations of senataxin function in disease pathogenesis and provides a novel example of allelic neurogenetic disorders with differing gene expression profiles. Weighted gene co-expression network analysis (WGCNA) demonstrated these senataxin-associated genes to be involved in both mutation-specific and shared functional gene networks. To assess this in vivo, we performed gene expression analysis on peripheral blood from members of 12 different AOA2 families and identified an AOA2-specific transcriptional signature. WGCNA identified two gene modules highly enriched for this transcriptional signature in the peripheral blood of all AOA2 patients studied. These modules were disease-specific and preserved in patient fibroblasts and in the cerebellum of Setx knockout mice demonstrating conservation across species and cell types, including neurons. These results identify novel genes and cellular pathways related to senataxin function in normal and disease states, and implicate alterations in gene expression as underlying the phenotypic differences between AOA2 and ALS4.

The Drosophila ortholog of MLL3 and MLL4, trithorax related, functions as a negative regulator of tissue growth.

The human MLL genes (MLL1 to MLL4) and their Drosophila orthologs, trithorax (trx) and trithorax related (trr), encode proteins capable of methylating histone H3 on lysine 4. MLL1 and MLL2 are most similar to trx, while MLL3 and MLL4 are more closely related to trr. Several MLL genes are mutated in human cancers, but how these proteins regulate cell proliferation is not known. Here we show that trr mutant cells have a growth advantage over their wild-type neighbors and display changes in the levels of multiple proteins that regulate growth and cell division, including Notch, Capicua, and cyclin B. trr mutant clones display markedly reduced levels of H3K4 monomethylation without obvious changes in the levels of H3K4 di- and trimethylation. The trr mutant phenotype resembles that of Utx, which encodes a H3K27 demethylase, consistent with the observation that Trr and Utx are found in the same protein complex. In contrast to the overgrowth displayed by trr mutant tissue, trx clones are underrepresented, express low levels of the antiapoptotic protein Diap1, and exhibit only modest changes in global levels of H3K4 methylation. Thus, in Drosophila eye imaginal discs, Trr, likely functioning together with Utx, restricts tissue growth. In contrast, Trx appears to promote cell survival.

Human-specific transcriptional networks in the brain.

Understanding human-specific patterns of brain gene expression and regulation can provide key insights into human brain evolution and speciation. Here, we use next-generation sequencing, and Illumina and Affymetrix microarray platforms, to compare the transcriptome of human, chimpanzee, and macaque telencephalon. Our analysis reveals a predominance of genes differentially expressed within human frontal lobe and a striking increase in transcriptional complexity specific to the human lineage in the frontal lobe. In contrast, caudate nucleus gene expression is highly conserved. We also identify gene coexpression signatures related to either neuronal processes or neuropsychiatric diseases, including a human-specific module with CLOCK as its hub gene and another module enriched for neuronal morphological processes and genes coexpressed with FOXP2, a gene important for language evolution. These data demonstrate that transcriptional networks have undergone evolutionary remodeling even within a given brain region, providing a window through which to view the foundation of uniquely human cognitive capacities.

RNASEQR--a streamlined and accurate RNA-seq sequence analysis program.

Next-generation sequencing (NGS) technologies-based transcriptomic profiling method often called RNA-seq has been widely used to study global gene expression, alternative exon usage, new exon discovery, novel transcriptional isoforms and genomic sequence variations. However, this technique also poses many biological and informatics challenges to extracting meaningful biological information. The RNA-seq data analysis is built on the foundation of high quality initial genome localization and alignment information for RNA-seq sequences. Toward this goal, we have developed RNASEQR to accurately and effectively map millions of RNA-seq sequences. We have systematically compared RNASEQR with four of the most widely used tools using a simulated data set created from the Consensus CDS project and two experimental RNA-seq data sets generated from a human glioblastoma patient. Our results showed that RNASEQR yields more accurate estimates for gene expression, complete gene structures and new transcript isoforms, as well as more accurate detection of single nucleotide variants (SNVs). RNASEQR analyzes raw data from RNA-seq experiments effectively and outputs results in a manner that is compatible with a wide variety of specialized downstream analyses on desktop computers.

Germline CBL mutations cause developmental abnormalities and predispose to juvenile myelomonocytic leukemia.

CBL encodes a member of the Cbl family of proteins, which functions as an E3 ubiquitin ligase. We describe a dominant developmental disorder resulting from germline missense CBL mutations, which is characterized by impaired growth, developmental delay, cryptorchidism and a predisposition to juvenile myelomonocytic leukemia (JMML). Some individuals experienced spontaneous regression of their JMML but developed vasculitis later in life. Importantly, JMML specimens from affected children show loss of the normal CBL allele through acquired isodisomy. Consistent with these genetic data, the common p.371Y>H altered Cbl protein induces cytokine-independent growth and constitutive phosphorylation of ERK, AKT and S6 only in hematopoietic cells in which normal Cbl expression is reduced by RNA interference. We conclude that germline CBL mutations have developmental, tumorigenic and functional consequences that resemble disorders that are caused by hyperactive Ras/Raf/MEK/ERK signaling and include neurofibromatosis type 1, Noonan syndrome, Costello syndrome, cardiofaciocutaneous syndrome and Legius syndrome.

A protocol for capturing daily variability in nursing care.

To meet current and future patient safety and quality requirements, traditional analyses based on data aggregated to the hospital or unit level over months or years may need to change. Nine customized databases were developed, five with patient data (e.g., age, illness severity, perceptions of nursing care quality, desired health outcomes) and four with nurse data (e.g., education, experience). These were merged to create a Patient-Nurse database. Nurse managers, clinicians, and researchers could use the protocol to conduct a more robust analysis of the relationships between nursing system characteristics and patient care processes and outcomes. The protocol described here could be used by nurse managers, clinicians, and researchers to better understand temporal phenomena and patient level data. Through information derived from applying this protocol, staffing decisions can be made to assure the right mix of nurses is available, not just the right number.

Mutations in CBL occur frequently in juvenile myelomonocytic leukemia.

Juvenile myelomonocytic leukemia is an aggressive myeloproliferative disorder characterized by malignant transformation in the hematopoietic stem cell compartment with proliferation of differentiated progeny. Seventy-five percent of patients harbor mutations in the NF1, NRAS, KRAS, or PTPN11 genes, which encode components of Ras signaling networks. Using single nucleotide polymorphism arrays, we identified a region of 11q isodisomy that contains the CBL gene in several JMML samples, and subsequently identified CBL mutations in 27 of 159 JMML samples. Thirteen of these mutations alter codon Y371. In this report, we also demonstrate that CBL and RAS/PTPN11 mutations were mutually exclusive in these patients. Moreover, the exclusivity of CBL mutations with respect to other Ras pathway-associated mutations indicates that CBL may have a role in deregulating this key pathway in JMML.

The largest prospective warfarin-treated cohort supports genetic forecasting.

Genetic variants of cytochrome P450 2C9 (CYP2C9) and vitamin K epoxide reductase (VKORC1) are known to influence warfarin dose, but the effect of other genes has not been fully elucidated. We genotyped 183 polymorphisms in 29 candidate genes in 1496 Swedish patients starting warfarin treatment, and tested for association with response. CYP2C9*2 and *3 explained 12% (P = 6.63 x 10(-34)) of the variation in warfarin dose, while a single VKORC1 SNP explained 30% (P = 9.82 x 10(-100)). No SNP outside the CYP2C gene cluster and VKORC1 regions was significantly associated with dose after correction for multiple testing. During initiation of therapy, homozygosity for CYP2C9 and VKORC1 variant alleles increased the risk of over-anticoagulation, hazard ratios 21.84 (95% CI 9.46; 50.42) and 4.56 (95% CI 2.85; 7.30), respectively. One of 8 patients with CYP2C9*3/*3 (12.5%) experienced severe bleeding during the first month compared with 0.27% of other patients (P = .066). A multiple regression model using the predictors CYP2C9, VKORC1, age, sex, and druginteractions explained 59% of the variance in warfarin dose, and 53% in an independent sample of 181 Swedish individuals. In conclusion, CYP2C9 and VKORC1 significantly influenced warfarin dose and predicted individuals predisposed to unstable anticoagulation. Our results strongly support that initiation of warfarin guided by pharmacogenetics would improve clinical outcome.

Use of a population-based survey to describe the health of Boston public housing residents.

OBJECTIVES: We compared the health of public housing residents with other Boston residents through a random-digit-dial survey. METHODS: We used data from the Boston Behavioral Risk Factor Surveillance System collected in 2001 and 2003 to make crude and demographically adjusted comparisons between public housing residents and other city residents on measures of health status, access and utilization, and health behaviors. RESULTS: Public housing residents were more likely to report fair or poor overall health status, ever-diagnosed hypertension, current asthma, ever-diagnosed diabetes, obesity, disability, loss of 6 or more teeth, and feelings of depression for 15 days or more in the past month. Public housing residents were slightly more likely than others to be without health insurance or report financial barriers to medical care. Public housing residents reported more smoking and physical inactivity, less past-month binge drinking and past-year marijuana use, and similar levels of lifetime drug use. CONCLUSIONS: Public housing residents reported substantially poorer health than did other city residents across a variety of conditions but similar levels of access to and utilization of health care. Public health departments may be able to use established surveys to measure health among public housing residents.

Association of warfarin dose with genes involved in its action and metabolism.

We report an extensive study of variability in genes encoding proteins that are believed to be involved in the action and biotransformation of warfarin. Warfarin is a commonly prescribed anticoagulant that is difficult to use because of the wide interindividual variation in dose requirements, the narrow therapeutic range and the risk of serious bleeding. We genotyped 201 patients for polymorphisms in 29 genes in the warfarin interactive pathways and tested them for association with dose requirement. In our study, polymorphisms in or flanking the genes VKORC1, CYP2C9, CYP2C18, CYP2C19, PROC, APOE, EPHX1, CALU, GGCX and ORM1-ORM2 and haplotypes of VKORC1, CYP2C9, CYP2C8, CYP2C19, PROC, F7, GGCX, PROZ, F9, NR1I2 and ORM1-ORM2 were associated with dose (P < 0.05). VKORC1, CYP2C9, CYP2C18 and CYP2C19 were significant after experiment-wise correction for multiple testing (P < 0.000175), however, the association of CYP2C18 and CYP2C19 was fully explained by linkage disequilibrium with CYP2C9*2 and/or *3. PROC and APOE were both significantly associated with dose after correction within each gene. A multiple regression model with VKORC1, CYP2C9, PROC and the non-genetic predictors age, bodyweight, drug interactions and indication for treatment jointly accounted for 62% of variance in warfarin dose. Weaker associations observed for other genes could explain up to approximately 10% additional dose variance, but require testing and validation in an independent and larger data set. Translation of this knowledge into clinical guidelines for warfarin prescription will be likely to have a major impact on the safety and efficacy of warfarin.

Single nucleotide polymorphism mapping using genome-wide unique sequences.

As more and more genomic DNAs are sequenced to characterize human genetic variations, the demand for a very fast and accurate method to genomically position these DNA sequences is high. We have developed a new mapping method that does not require sequence alignment. In this method, we first identified DNA fragments of 15 bp in length that are unique in the human genome and then used them to position single nucleotide polymorphism (SNP) sequences. By use of four desktop personal computers with AMD K7 (1 GHz) processors, our new method mapped more than 1.6 million SNP sequences in 20 hr and achieved a very good agreement with mapping results from alignment-based methods.