Minimum redundancy maximum relevance feature selection approach for temporal gene expression data

BMC Bioinformatics, Journal Year: 2017, Volume and Issue: 18(1)

Published: Jan. 3, 2017

Feature selection, aiming to identify a subset of features among possibly large set that are relevant for predicting response, is an important preprocessing step in machine learning. In gene expression studies this not trivial task several reasons, including potential temporal character data. However, most feature selection approaches developed microarray data cannot handle multivariate without previous flattening, which results loss information. We propose minimum redundancy - maximum relevance (TMRMR) approach, able flattening. the proposed approach we compute by averaging F-statistic values calculated across individual time steps, and between genes using dynamical warping approach. The method evaluated on three datasets from human viral challenge studies. Obtained show outperforms alternatives widely used particular, achieved improvement accuracy 34 out 54 experiments, while other methods outperformed it no more than 4 experiments. filter-based based criteria. incorporates information combining relevance, as average value different with redundancy, employing As evident our incorporating into process leads discriminative features.

Language: Английский

---

A variable selection method for genome-wide association studies

Bioinformatics, Journal Year: 2010, Volume and Issue: 27(1), P. 1 - 8

Published: Oct. 29, 2010

Abstract Motivation: Genome-wide association studies (GWAS) involving half a million or more single nucleotide polymorphisms (SNPs) allow genetic dissection of complex diseases in holistic manner. The common practice analyzing one SNP at time does not fully realize the potential GWAS to identify multiple causal variants and predict risk disease. Existing methods for joint analysis data tend miss SNPs that are marginally uncorrelated with disease have high false discovery rates (FDRs). Results: We introduce GWASelect, statistically powerful computationally efficient variable selection method designed tackle unique challenges data. This searches iteratively over conditional on previously selected is thus capable capturing correlated as well those A special resampling mechanism built into reduce positive findings. Simulation demonstrate GWASelect performs under wide spectrum linkage disequilibrium patterns can be substantially than existing while having lower FDR. In addition, regression models based yield accurate prediction methods. advantages illustrated Wellcome Trust Case-Control Consortium (WTCCC) Availability: software implementing available http://www.bios.unc.edu/~lin. Access WTCCC data: http://www.wtccc.org.uk/ Contact: [email protected] Supplementary information: Bioinformatics Online.

Language: Английский

---

A survey about methods dedicated to epistasis detection

Frontiers in Genetics, Journal Year: 2015, Volume and Issue: 6

Published: Sept. 10, 2015

During the past decade, findings of genome-wide association studies (GWAS) improved our knowledge and understanding disease genetics. To date, thousands SNPs have been associated to diseases other complex traits. Statistical analysis typically looks for between a phenotype SNP taken individually via single-locus tests. However, geneticists admit this is an oversimplified approach tackle complexity underlying biological mechanisms. Interaction SNPs, namely epistasis, must be considered. Unfortunately, epistasis detection gives rise analytic challenges since analyzing every combination at present impractical scale. In review, we will main strategies recently proposed detect epistatic interactions, along with their operating principle. Some these methods are exhaustive, such as multifactor dimensionality reduction, likelihood ratio-based tests or receiver characteristic curve analysis; some non-exhaustive, machine learning techniques (random forests, Bayesian networks) combinatorial optimization approaches (ant colony optimization, computational evolution system).

Language: Английский

---

A century after Fisher: time for a new paradigm in quantitative genetics

Trends in Genetics, Journal Year: 2013, Volume and Issue: 29(12), P. 669 - 676

Published: Oct. 23, 2013

Language: Английский

---

Genomes of Novel Microbial Lineages Assembled from the Sub-Ice Waters of Lake Baikal

Applied and Environmental Microbiology, Journal Year: 2017, Volume and Issue: 84(1)

Published: Oct. 27, 2017

ABSTRACT We present a metagenomic study of Lake Baikal (East Siberia). Two samples obtained from the water column under ice cover (5 and 20 m deep) in March 2016 have been deep sequenced reads assembled to generate metagenome-assembled genomes (MAGs) that are representative microbes living this special environment. Compared with freshwater bodies studied around world, had an unusually high fraction Verrucomicrobia . Other groups, such as Actinobacteria Proteobacteria , were proportions similar those found other lakes. The (and probably cells) tended be small, presumably reflecting extremely oligotrophic cold prevalent conditions. novel lineages recruiting very little distantly related microbes. Despite their novelty, they showed closest relationship discovered by approaches lakes reservoirs. Some them particularly MAGs Baltic Sea, which, although it is brackish, connected ocean, much more eutrophic, has climatological Many contained rhodopsin genes, indicating that, spite decreased light penetration allowed thick ice/snow cover, photoheterotrophy could widespread column, either because enough penetrates or already adapted summer ice-less SAR11 subtype I/II showing striking synteny Pelagibacter ubique strains, well phage infecting bacterium Polynucleobacter IMPORTANCE increasing number studies on different bodies, there still missing component suffering long seasonal frozen cycles. Here, we describe microbial assemblies appear upper Baikal, largest deepest body Earth. This lake January May, which generates conditions include inverted temperature gradient (colder up), decrease due ice, and, especially, snow open-ocean high-altitude than brackish systems. As expected, most reconstructed others environments, like Sea Among them, was broad set streamlined small genomes/intergenic spacers, including new nonmarine -like (subtype I/II) genome.

Language: Английский

---

Incorporating deep learning with convolutional neural networks and position specific scoring matrices for identifying electron transport proteins

Journal of Computational Chemistry, Journal Year: 2017, Volume and Issue: 38(23), P. 2000 - 2006

Published: June 22, 2017

In several years, deep learning is a modern machine technique using in variety of fields with state‐of‐the‐art performance. Therefore, utilization to enhance performance also an important solution for current bioinformatics field. this study, we try use via convolutional neural networks and position specific scoring matrices identify electron transport proteins, which molecular function transmembrane proteins. Our method can approach precise model identifying proteins achieved sensitivity 80.3%, specificity 94.4%, accuracy 92.3%, MCC 0.71 independent dataset. The proposed serve as powerful tool help biologists understand the Moreover, study provides basis further research that enrich field applying bioinformatics. © 2017 Wiley Periodicals, Inc.

Language: Английский

---

Comparative and functional genomics of the Lactococcus lactis taxon; insights into evolution and niche adaptation

BMC Genomics, Journal Year: 2017, Volume and Issue: 18(1)

Published: March 29, 2017

Lactococcus lactis is among the most widely studied lactic acid bacterial species due to its long history of safe use and economic importance dairy industry, where it exploited as a starter culture in cheese production.In current study, we report on complete sequencing 16 L. subsp. cremoris genomes. The chromosomal features these strains conjunction with 14 completely sequenced, publicly available lactococcal chromosomes were assessed particular emphasis discerning subspecies division, evolution niche adaptation. deduced pan-genome was found be closed, indicating that representative data sets employed for this analysis are sufficient fully describe genetic diversity taxon.Niche adaptation appears play significant role governing content each subspecies, while (differential) genome decay redundancy also highlighted.

Language: Английский

---

Tracking microbial colonization in fecal microbiota transplantation experiments via genome-resolved metagenomics

Microbiome, Journal Year: 2017, Volume and Issue: 5(1)

Published: May 4, 2017

Fecal microbiota transplantation (FMT) is an effective treatment for recurrent Clostridium difficile infection and shows promise treating other medical conditions associated with intestinal dysbioses. However, we lack a sufficient understanding of which microbial populations successfully colonize the recipient gut, widely used approaches to study ecology FMT experiments fail provide enough resolution identify that are likely responsible FMT-derived benefits. We shotgun metagenomics together assembly binning strategies reconstruct metagenome-assembled genomes (MAGs) from fecal samples single donor. then metagenomic mapping track occurrence distribution patterns donor MAGs in two recipients. Our analyses revealed 22% 92 highly complete bacterial identified colonized remained abundant recipients at least 8 weeks. Most high colonization rate belonged order Bacteroidales. The vast majority those lacked evidence Clostridiales, success was negatively correlated number genes related sporulation. analysis 151 publicly available gut metagenomes showed both were prevalent, ones neither rare across participants Human Microbiome Project. Although our dataset link between taxonomy ability given MAG, also belong same taxon different properties, highlighting importance appropriate level explore functional basis targets cultivation, hypothesis generation, testing model systems. analytical strategy adopted can genomic insights into may be critical efficacy due their metabolic guide cultivation efforts investigate mechanistic underpinnings this procedure beyond associations.

Language: Английский

---

Using Bayesian networks to discover relations between genes, environment, and disease

BioData Mining, Journal Year: 2013, Volume and Issue: 6(1)

Published: March 21, 2013

We review the applicability of Bayesian networks (BNs) for discovering relations between genes, environment, and disease. By translating probabilistic dependencies among variables into graphical models vice versa, BNs provide a comprehensible modular framework representing complex systems. first describe network approach its to understanding genetic environmental basis then variety algorithms learning structure from observational data. Because their relevance real-world applications, topics missing data causal interpretation are emphasized. The BN is exemplified through application population-based study bladder cancer in New Hampshire, USA. For didactical purposes, we intentionally keep this example simple. When applied complete records, find only minor differences performance results different algorithms. Subsequent incorporation partial records EM algorithm gives us greater power detect relations. Allowing structures that depart strict also enhances our ability discover associations including gene-gene (epistasis) gene-environment interactions. While already powerful tools dissection disease generation prognostic models, there remain some conceptual computational challenges. These include proper handling continuous unmeasured factors, explicit prior knowledge, evaluation communication robustness substantive conclusions alternative assumptions manifestations.

Language: Английский

---

Learning genetic epistasis using Bayesian network scoring criteria

BMC Bioinformatics, Journal Year: 2011, Volume and Issue: 12(1)

Published: March 31, 2011

Gene-gene epistatic interactions likely play an important role in the genetic basis of many common diseases. Recently, machine-learning and data mining methods have been developed for learning relationships from data. A well-known combinatorial method that has successfully applied detecting epistasis is Multifactor Dimensionality Reduction (MDR). Jiang et al. created a called BNMBL to learn Bayesian network (BN) models. They compared MDR using simulated sets. Each these sets was generated model associates two SNPs with disease includes 18 unrelated SNPs. For each set, were used score all 2-SNP models, learned significantly more correct In real sets, we ordinarily do not know number influence phenotype. may perform as well if also scored models containing than Furthermore, other BN scoring criteria developed. detect even better BNMBL. Although BNs are promising tool data, cannot confidently use them this domain until determine which work best or when try without knowledge model. We evaluated performance 22 28,000 Alzheimer's GWAS set. Our results surprising criterion large values hyperparameter α performed best. This at recall hardest-to-detect substantiating previous conclude representing holds promise identifying variants particular, appears alternatives.

Language: Английский

---