Domain Adaptation with Conditional Transferable Components.
Ontology highlight
ABSTRACT:
SUBMITTER: Gong M
PROVIDER: S-EPMC5321138 | biostudies-literature | 2016 Jun
REPOSITORIES: biostudies-literature
Publications
Domain Adaptation with Conditional Transferable Components.
Gong Mingming M Zhang Kun K Liu Tongliang T Tao Dacheng D Glymour Clark C Schölkopf Bernhard B
JMLR workshop and conference proceedings 20160601
Domain adaptation arises in supervised learning when the training (source domain) and test (target domain) data have different distributions. Let <i>X</i> and <i>Y</i> denote the features and target, respectively, previous work on domain adaptation mainly considers the covariate shift situation where the distribution of the features <i>P</i>(<i>X</i>) changes across domains while the conditional distribution <i>P</i>(<i>Y</i>∣<i>X</i>) stays the same. To reduce domain discrepancy, recent methods ...[more]
Similar Datasets
Project description:Divergent natural selection promotes local adaptation and can lead to reproductive isolation of populations in contrasting environments; however, the genetic basis of local adaptation remains largely unresolved in natural populations. Local adaptation might result from antagonistic pleiotropy, where alternate alleles are favoured in distinct habitats, and polymorphism is maintained by selection. Alternatively, under conditional neutrality some alleles may be favoured in one environment but neutral at other locations. Antagonistic pleiotropy maintains genetic variation across the landscape; however, there is a systematic bias against discovery of antagonistic pleiotropy because the fitness benefits of local alleles need to be significant in at least two environments. Here, we develop a generally applicable method to investigate polygenic local adaptation and identify loci that are the targets of selection. This approach evaluates allele frequency changes after selection at loci across the genome to distinguish antagonistic pleiotropy from conditional neutrality and deleterious variation. We investigate local adaptation at the qualitative trait loci (QTL) level in field experiments, in which we expose 177 F(6) recombinant inbred lines and parental lines of Boechera stricta (Brassicaceae) to their parental environments over two seasons. We demonstrate polygenic selection for native alleles in both environments, with 2.8% of the genome exhibiting antagonistic pleiotropy and 8% displaying conditional neutrality. Our study strongly supports antagonistic pleiotropy at one large-effect flowering phenology QTL (nFT): native homozygotes had significantly greater probabilities of flowering than foreign homozygotes in both parental environments. Such large-scale field studies are essential to elucidate the genetic basis of adaptation in natural populations.
| S-EPMC3492549 | biostudies-literature
Project description:Proteins are often classified in a binary fashion as either structured or disordered. However this approach has several deficits. Firstly, protein folding is always conditional on the physiochemical environment. A protein which is structured in some circumstances will be disordered in others. Secondly, it hides a fundamental asymmetry in behavior. While all structured proteins can be unfolded through a change in environment, not all disordered proteins have the capacity for folding. Failure to accommodate these complexities confuses the definition of both protein structural domains and intrinsically disordered regions. We illustrate these points with an experimental study of a family of small binding domains, drawn from the RNA polymerase of mumps virus and its closest relatives. Assessed at face value the domains fall on a structural continuum, with folded, partially folded, and near unstructured members. Yet the disorder present in the family is conditional, and these closely related polypeptides can access the same folded state under appropriate conditions. Any heuristic definition of the protein domain emphasizing conformational stability divides this domain family in two, in a way that makes no biological sense. Structural domains would be better defined by their ability to adopt a specific tertiary structure: a structure that may or may not be realized, dependent on the circumstances. This explicitly allows for the conditional nature of protein folding, and more clearly demarcates structural domains from intrinsically disordered regions that may function without folding.
| S-EPMC3831666 | biostudies-literature
Project description:Domain adaptation aims to correct the classifiers when faced with distribution shift between source (training) and target (test) domains. State-of-the-art domain adaptation methods make use of deep networks to extract domain-invariant representations. However, existing methods assume that all the instances in the source domain are correctly labeled; while in reality, it is unsurprising that we may obtain a source domain with noisy labels. In this paper, we are the first to comprehensively investigate how label noise could adversely affect existing domain adaptation methods in various scenarios. Further, we theoretically prove that there exists a method that can essentially reduce the side-effect of noisy source labels in domain adaptation. Specifically, focusing on the generalized target shift scenario, where both label distribution PY and the class-conditional distribution P X|Y can change, we discover that the denoising Conditional Invariant Component (DCIC) framework can provably ensures (1) extracting invariant representations given examples with noisy labels in the source domain and unlabeled examples in the target domain and (2) estimating the label distribution in the target domain with no bias. Experimental results on both synthetic and real-world data verify the effectiveness of the proposed method.
| S-EPMC8043624 | biostudies-literature
Project description:The multidrug resistance-encoding IncA/C conjugative plasmids disseminate antibiotic resistance genes among clinically relevant enteric bacteria. A plasmid-encoded disulfide isomerase is associated with conjugation. Sequence analysis of several IncA/C plasmids and IncA/C-related integrative and conjugative elements (ICE) from commensal and pathogenic bacteria identified a conserved DsbC/DsbG homolog (DsbP). The crystal structure of DsbP reveals an N-terminal domain, a linker region, and a C-terminal catalytic domain. A DsbP homodimer is formed through domain swapping of two DsbP N-terminal domains. The catalytic domain incorporates a thioredoxin-fold with characteristic CXXC and cis-Pro motifs. Overall, the structure and redox properties of DsbP diverge from the Escherichia coli DsbC and DsbG disulfide isomerases. Specifically, the V-shaped dimer of DsbP is inverted compared with EcDsbC and EcDsbG. In addition, the redox potential of DsbP (-161 mV) is more reducing than EcDsbC (-130 mV) and EcDsbG (-126 mV). Other catalytic properties of DsbP more closely resemble those of EcDsbG than EcDsbC. These catalytic differences are in part a consequence of the unusual active site motif of DsbP (CAVC); substitution to the EcDsbC-like (CGYC) motif converts the catalytic properties to those of EcDsbC. Structural comparison of the 12 independent subunit structures of DsbP that we determined revealed that conformational changes in the linker region contribute to mobility of the catalytic domain, providing mechanistic insight into DsbP function. In summary, our data reveal that the conserved plasmid-encoded DsbP protein is a bona fide disulfide isomerase and suggest that a dedicated oxidative folding enzyme is important for conjugative plasmid transfer.
| S-EPMC3908391 | biostudies-literature
Project description:Strain inoculation (bioaugmentation) is a potentially useful technology to provide microbiomes with new functionalities. However, there is limited understanding of the genetic factors contributing to successful establishment of inoculants. This work aimed to characterize the genes implicated in proliferation of the monoaromatic compound-degrading Pseudomonas veronii 1YdBTEX2 in nonsterile polluted soils. We generated two independent mutant libraries by random minitransposon-delivered marker insertion followed by deep sequencing (Tn-seq) with a total of 5.0 × 105 unique insertions. Libraries were grown in multiple successive cycles for up to 50 generations either in batch liquid medium or in two types of soil microcosms with different resident microbial content (sand or silt) in the presence of toluene. Analysis of gene insertion abundances at different time points (passed generations of metapopulation growth), in comparison to proportions at start and to in silico generated randomized insertion distributions, allowed to define ~800 essential genes common to both libraries and ~2,700 genes with conditional fitness effects in either liquid or soil (195 of which resulted in fitness gain). Conditional fitness genes largely overlapped among all growth conditions but affected approximately twice as many functions in liquid than in soil. This indicates soil to be a more promiscuous environment for mutant growth, probably because of additional nutrient availability. Commonly depleted genes covered a wide range of biological functions and metabolic pathways, such as inorganic ion transport, fatty acid metabolism, amino acid biosynthesis, or nucleotide and cofactor metabolism. Only sparse gene sets were uncovered whose insertion caused fitness decrease exclusive for soils, which were different between silt and sand. Despite detectable higher resident bacteria and potential protist predatory counts in silt, we were, therefore, unable to detect any immediately obvious candidate genes affecting P. veronii biological competitiveness. In contrast to liquid growth conditions, mutants inactivating flagella biosynthesis and motility consistently gained strong fitness advantage in soils and displayed higher growth rates than wild type. In conclusion, although many gene functions were found to be important for growth in soils, most of these are not specific as they affect growth in liquid minimal medium more in general. This indicates that P. veronii does not need major metabolic reprogramming for proliferation in soil with accessible carbon and generally favorable growth conditions. IMPORTANCE Restoring damaged microbiomes is still a formidable challenge. Classical widely adopted approaches consist of augmenting communities with pure or mixed cultures in the hope that these display their intended selected properties under in situ conditions. Ecological theory, however, dictates that introduction of a nonresident microbe is unlikely to lead to its successful proliferation in a foreign system such as a soil microbiome. In an effort to study this systematically, we used random transposon insertion scanning to identify genes and possibly, metabolic subsystems, that are crucial for growth and survival of a bacterial inoculant (Pseudomonas veronii) for targeted degradation of monoaromatic compounds in contaminated nonsterile soils. Our results indicate that although many gene functions are important for proliferation in soil, they are general factors for growth and not exclusive for soil. In other words, P. veronii is a generalist that is not a priori hindered by the soil for its proliferation and would make a good bioaugmentation candidate.
| S-EPMC10134887 | biostudies-literature
Project description:MotivationDomain adaptation allows for the development of predictive models even in cases with limited sample data. Weighted elastic net domain adaptation specifically leverages features of genomic data to maximize transferability but the method is too computationally demanding to apply to many genome-sized datasets.ResultsWe developed wenda_gpu, which uses GPyTorch to train models on genomic data within hours on a single GPU-enabled machine. We show that wenda_gpu returns comparable results to the original wenda implementation, and that it can be used for improved prediction of cancer mutation status on small sample sizes than regular elastic net.Availability and implementationwenda_gpu is available on GitHub at https://github.com/greenelab/wenda_gpu/.Supplementary informationSupplementary data are available at Bioinformatics online.
| S-EPMC9665854 | biostudies-literature
Project description:In this study, we examine the feasibility and limitations of describing the motional behavior of three-domain proteins in which the domains are linearly connected. In addition to attempting the determination of the internal and overall reorientational correlation times, we investigate the existence of correlations in the motions between the three domains. Since in linearly arranged three-domain proteins, there are typically no experimental data that can directly report on motional correlation between the first and the third domain, we address this question by dynamics simulations. Two limiting cases occur: (1) for weak repulsive potentials and (2) when strong repulsive potentials are applied between sequential domains. The motions of the terminal domains become correlated in the case of strong interdomain repulsive potentials when these potentials do not allow the angle between the sequential domains to be smaller than about 60°. Using the model-free (MF) and extended MF formalisms of Lipari and Szabo, we find that the motional behavior can be separated into two components; the first component represents the concerted overall motion of the three domains, and the second describes the independent component of the motion of each individual domain. We find that this division of the motional behavior of the protein is maintained only when their timescales are distinct and can be made when the angles between sequential domains remain between 60° and 160°. In this work, we identify and quantify interdomain motional correlations.
| S-EPMC4181349 | biostudies-literature
Project description:A cardiac hypertrophy is defined as an increase in heart mass which may either be beneficial (physiological hypertrophy) or detrimental (pathological hypertrophy). This study was undertaken to establish the role of different protein kinase-C (PKC) isoforms in the regulation of cardiac adaptation during two types of cardiac hypertrophy. Phosphorylation of specific PKC-isoforms and expression of their downstream proteins were studied during physiological and pathological hypertrophy in 24 week male Balb/c mice (Mus musculus) models, by reverse transcriptase-PCR, western blot analysis and M-mode echocardiography for cardiac function analysis. PKC-δ was significantly induced during pathological hypertrophy while PKC-α was exclusively activated during physiological hypertrophy in our study. PKC-δ activation during pathological hypertrophy resulted in cardiomyocyte apoptosis leading to compromised cardiac function and on the other hand, activation of PKC-α during physiological hypertrophy promoted cardiomyocyte growth but down regulated cellular apoptotic load resulting in improved cardiac function. Reversal in PKC-isoform with induced activation of PKC-δ and simultaneous inhibition of phospho-PKC-α resulted in an efficient myocardium to deteriorate considerably resulting in compromised cardiac function during physiological hypertrophy via augmentation of apoptotic and fibrotic load. This is the first report where PKC-α and -δ have been shown to play crucial role in cardiac adaptation during physiological and pathological hypertrophy respectively thereby rendering compromised cardiac function to an otherwise efficient heart by conditional reversal of their activation.
| S-EPMC4130596 | biostudies-literature
Project description:In vivo, the human genome folds into a characteristic ensemble of 3D structures. The mechanism driving the folding process remains unknown. We report a theoretical model for chromatin (Minimal Chromatin Model) that explains the folding of interphase chromosomes and generates chromosome conformations consistent with experimental data. The energy landscape of the model was derived by using the maximum entropy principle and relies on two experimentally derived inputs: a classification of loci into chromatin types and a catalog of the positions of chromatin loops. First, we trained our energy function using the Hi-C contact map of chromosome 10 from human GM12878 lymphoblastoid cells. Then, we used the model to perform molecular dynamics simulations producing an ensemble of 3D structures for all GM12878 autosomes. Finally, we used these 3D structures to generate contact maps. We found that simulated contact maps closely agree with experimental results for all GM12878 autosomes. The ensemble of structures resulting from these simulations exhibited unknotted chromosomes, phase separation of chromatin types, and a tendency for open chromatin to lie at the periphery of chromosome territories.
| S-EPMC5087044 | biostudies-literature
Project description:Produce is increasingly recognized as a reservoir of human pathogens and transferable antibiotic resistance genes. This study aimed to explore methods to characterize the transferable resistome of bacteria associated with produce. Mixed salad, arugula, and cilantro purchased from supermarkets in Germany were analyzed by means of cultivation- and DNA-based methods. Before and after a nonselective enrichment step, tetracycline (TET)-resistant Escherichia coli were isolated and plasmids conferring TET resistance were captured by exogenous plasmid isolation. TET-resistant E. coli isolates, transconjugants, and total community DNA (TC-DNA) from the microbial fraction detached from leaves or after enrichment were analyzed for the presence of resistance genes, class 1 integrons, and various plasmids by real-time PCR and PCR-Southern blot hybridization. Real-time PCR primers were developed for IncI and IncF plasmids. TET-resistant E. coli isolated from arugula and cilantro carried IncF, IncI1, IncN, IncHI1, IncU, and IncX1 plasmids. Three isolates from cilantro were positive for IncN plasmids and blaCTX-M-1 From mixed salad and cilantro, IncF, IncI1, and IncP-1β plasmids were captured exogenously. Importantly, whereas direct detection of IncI and IncF plasmids in TC-DNA failed, these plasmids became detectable in DNA extracted from enrichment cultures. This confirms that cultivation-independent DNA-based methods are not always sufficiently sensitive to detect the transferable resistome in the rare microbiome. In summary, this study showed that an impressive diversity of self-transmissible multiple resistance plasmids was detected in bacteria associated with produce that is consumed raw, and exogenous capturing into E. coli suggests that they could transfer to gut bacteria as well.IMPORTANCE Produce is one of the most popular food commodities. Unfortunately, leafy greens can be a reservoir of transferable antibiotic resistance genes. We found that IncF and IncI plasmids were the most prevalent plasmid types in E. coli isolates from produce. This study highlights the importance of the rare microbiome associated with produce as a source of antibiotic resistance genes that might escape cultivation-independent detection, yet may be transferred to human pathogens or commensals.
| S-EPMC6222124 | biostudies-literature