Comprehensible Predictive Modeling Using Regularized Logistic Regression and Comorbidity Based Features

Different studies have demonstrated the importance of comorbidities to better understand the origin and evolution of medical complications. This study focuses on improvement of the predictive model interpretability based on simple logical features representing comorbidities. We use group lasso based feature interaction discovery followed by a post-processing step, where simple logic terms are added. In the final step, we reduce the feature set by applying lasso logistic regression to obtain a compact set of non-zero coefficients that represent a more comprehensible predictive model. The effectiveness of the proposed approach was demonstrated on a pediatric hospital discharge dataset that was used to build a readmission risk estimation model. The evaluation of the proposed method demonstrates a reduction of the initial set of features in a regression model by 72%, with a slight improvement in the Area Under the ROC Curve metric from 0.763 (95% CI: 0.755–0.771) to 0.769 (95% CI: 0.761–0.777). Additionally, our results show improvement in comprehensibility of the final predictive model using simple comorbidity based terms for logistic regression.     

Synthesis of Isoxazolidino Analogues of 2′,3′-Dideoxynucleosides

Abstract

The complete set of the 4′-aza analogues of 2′,3′-dideoxynucleosides was synthesized by cycloaddition of N-tetrahydropiranyl or N-trityl methylene nitrones on suitably protected vinyl nucleobases. The convertible nucleoside approach was used in the preparation of cytosine and 5-methyl cytosine analogues.     

Phylogenomics of the Maverick Virus-Like Mobile Genetic Elements of Vertebrates

Mavericks are virus-like mobile genetic elements found in the genomes of eukaryotes. Although Mavericks encode capsid morphogenesis homologs, their viral particles have not been observed. Here, we provide new evidence supporting the viral nature of Mavericks and the potential existence of virions. To this end, we conducted a phylogenomic analysis of Mavericks in hundreds of vertebrate genomes, discovering 134 elements with an intact coding capacity in 17 host species. We reveal an extensive genomic fossil record in 143 species and date three groups of elements to the Late Cretaceous. Bayesian phylogenetic analysis using genomic fossil orthologs suggests that Mavericks have infected osteichthyans for ∼419 My. They have undergone frequent cross-species transmissions in cyprinid fish and all core genes are subject to strong purifying selection. We conclude that vertebrate Mavericks form an ancient lineage of aquatic dsDNA viruses which are probably still functional in some vertebrate lineages.     

BODIPY for photodynamic therapy applications: computational study of the effect of bromine substitution on <Superscript>1</Superscript>O<Subscript>2</Subscript> photosensitization

Density functional theory and its time-dependent extension (DFT, TDDFT) were employed to establish the feasibility of using a series of 4,4-difluoro-4-bora-3a,4a-diaza-s-indacenes (BODIPYs) in photodynamic therapy. Their absorption electronic spectra, singlet–triplet energy gaps, and spin–orbit matrix elements were computed and are discussed here. The effects of bromine substitution on the photophysical properties of BODIPY were elucidated. The investigated compounds were found to possess different excited triplet states that lie below the energy of the bright excited singlet state (S₁ or S₂), depending on the positions occupied by the bromine atoms. The computed spin–orbit matrix elements for the radiationless intersystem crossing S_n?→ ?T_m and the relative singlet–triplet energy gaps allowed the prediction of plausible nonradiative decay pathways for the production of singlet excited molecular oxygen, the key cytotoxic agent in photodynamic therapy.

Open image in new window

Graphical Abstract The photophysical properties affected by the presence of bromine atoms in different positions of a BODIPY core have been here elucidated. In particular it has been found that SOC values strongly depend on the position of heavy atoms into the BODIPY core, suggesting positions 1 and 7 as the best ones to enhance the ISC kinetics

     

Synthesis of isoxazolidino analogues of 2',3'-dideoxynucleosides.

The complete set of the 4'-aza analogues of 2',3'-dideoxynucleosides was synthesized by cycloaddition of N-tetrahydropiranyl or N-trityl methylene nitrones on suitably protected vinyl nucleobases. The convertible nucleoside approach was used in the preparation of cytosine and 5-methyl cytosine analogues.     

Lignocellulose-degrading enzyme production by white-rot <Emphasis Type="Italic">Basidiomycetes</Emphasis> isolated from the forests of Georgia

The production of lignocellulolytic enzymes by eleven basidiomycetes species isolated from two ecosystems of Georgia was investigated for the first time under submerged (SF) and solid-state fermentation (SSF) of lignocellulosic by-products. Notable intergeneric and intrageneric differences were revealed with regard to the extent of hydrolase and oxidase activity. Several fungi produced laccase along with hydrolases in parallel with growth during the trophophase, showing that the synthesis of this enzyme is not connected with secondary metabolism. The lignocellulosic substrate type had the greatest impact on enzyme secretion. Some of the substrates significantly stimulated lignocellulolytic enzyme synthesis without supplementation of the culture medium with specific inducers. Exceptionally high carboxymethyl cellulase (CMCase, 122 U ml⁻¹) and xylanase (195 U ml⁻¹) activities were revealed in SF of mandarin peelings by Pseudotremella gibbosa IBB 22 and of residue after ethanol production (REP) by Fomes fomentarius IBB 38, respectively. The SSF of REP by T. pubescens IBB 11 ensured the highest level of laccase activity (24,690 U l⁻¹), whereas the SSF of wheat bran and SF of mandarin peels provided the highest manganese peroxidase activity (570–620 U l⁻¹) of Trichaptum biforme IBB 117. Moreover, the variation of lignocellulosic growth substrate provides an opportunity to obtain enzyme preparations containing different ratios of individual enzymes.     

Enhanced gene amplification in human cells knocked down for DNA-PKcs

Gene amplification, a key mechanism for oncogene activation and drug resistance in tumour cells, involves the generation and joining of DNA double-strand breaks. Amplified DNA can be carried either on intra-chromosomal arrays or on extra-chromosomal elements (double minutes). We previously showed that, in rodent cells deficient in DNA-PKcs, intra-chromosomal amplification is significantly enhanced. In the present work, we studied gene amplification in human HeLa cell lines in which the expression of the DNA-PKcs gene was constitutively inhibited by shRNAs. These cell lines showed an increased sensitivity to ionizing radiations, an enhanced frequency of chromosomal aberrations and an increased rate of occurrence of methotrexate resistant colonies compared to the control cell lines (6-18 times). The main mechanism of resistance to methotrexate was extra-chromosomal amplification of the dihydrofolate reductase gene. These results indicate that, in human cells, inhibition of DNA-PKcs gene expression favours gene amplification occurring via the production of double minutes. In addition, they show that cell lines constitutively expressing shRNAs are good model systems to study the role of specific functions in gene amplification.     

Ambient Air Pollution and the Progression of Atherosclerosis in Adults

Background

Cross-sectional studies suggest an association between exposure to ambient air pollution and atherosclerosis. We investigated the association between outdoor air quality and progression of subclinical atherosclerosis (common carotid artery intima-media thickness, CIMT).

Methodology/Principal Findings

We examined data from five double-blind randomized trials that assessed effects of various treatments on the change in CIMT. The trials were conducted in the Los Angeles area. Spatial models and land-use data were used to estimate the home outdoor mean concentration of particulate matter up to 2.5 micrometer in diameter (PM2.5), and to classify residence by proximity to traffic-related pollution (within 100 m of highways). PM2.5 and traffic proximity were positively associated with CIMT progression. Adjusted coefficients were larger than crude associations, not sensitive to modelling specifications, and statistically significant for highway proximity while of borderline significance for PM2.5 (P = 0.08). Annual CIMT progression among those living within 100 m of a highway was accelerated (5.5 micrometers/yr [95%CI: 0.13–10.79; p = 0.04]) or more than twice the population mean progression. For PM2.5, coefficients were positive as well, reaching statistical significance in the socially disadvantaged; in subjects reporting lipid lowering treatment at baseline; among participants receiving on-trial treatments; and among the pool of four out of the five trials.

Conclusion

Consistent with cross-sectional findings and animal studies, this is the first study to report an association between exposure to air pollution and the progression of atherosclerosis – indicated with CIMT change – in humans. Ostensibly, our results suggest that air pollution may contribute to the acceleration of cardiovascular disease development – the main causes of morbidity and mortality in many countries. However, the heterogeneity of the volunteering populations across the five trials, the limited sample size within trials and other relevant subgroups, and the fact that some key findings reached statistical significance in subgroups rather than the sample precludes generalizations to the general population.     

Label-Free Protein Quantification for Plant Golgi Protein Localization and Abundance

The proteomic composition of the Arabidopsis (Arabidopsis thaliana) Golgi apparatus is currently reasonably well documented; however, little is known about the relative abundances between different proteins within this compartment. Accurate quantitative information of Golgi resident proteins is of great importance: it facilitates a better understanding of the biochemical processes that take place within this organelle, especially those of different polysaccharide synthesis pathways. Golgi resident proteins are challenging to quantify because the abundance of this organelle is relatively low within the cell. In this study, an organelle fractionation approach targeting the Golgi apparatus was combined with a label-free quantitative mass spectrometry (data-independent acquisition method using ion mobility separation known as LC-IMS-MS^E [or HDMS^E]) to simultaneously localize proteins to the Golgi apparatus and assess their relative quantity. In total, 102 Golgi-localized proteins were quantified. These data show that organelle fractionation in conjunction with label-free quantitative mass spectrometry is a powerful and relatively simple tool to access protein organelle localization and their relative abundances. The findings presented open a unique view on the organization of the plant Golgi apparatus, leading toward unique hypotheses centered on the biochemical processes of this organelle.The plant Golgi apparatus plays an important role in protein and lipid glycosylation and sorting as well as biosynthesis of large amounts of extracellular polysaccharides. It contains a large and diverse set of glycosyltransferases and other enzymes that are required for the synthesis and modification of these polysaccharides (Parsons et al., 2012b; Oikawa et al., 2013). The protein composition of this organelle has been the focus of a number of studies; however, these studies largely report a catalog of Golgi-localized proteins, and to date, there are no comprehensive data on the relative abundance of the different protein constituents of the Golgi apparatus (Dunkley et al., 2004, 2006; Sadowski et al., 2008; Nikolovski et al., 2012; Groen et al., 2014). The quantification of the plant Golgi proteome has been considered challenging, because this organelle is proportionally of low abundance in the cell; therefore, its constituent proteins are rarely identified in conventional proteomics experiments. Investigation of such low-abundance proteins generally requires sample fractionation on the organelle, protein, or peptide level (Stasyk and Huber, 2004; Haynes and Roberts, 2007; Di Palma et al., 2012).Here, an organelle fractionation approach in conjunction with label-free quantitative proteomic analysis was used to assess the localization and relative abundance of proteins within the plant Golgi apparatus. Label-free quantification is an increasingly popular alternative to isotopic tagging quantitative methods; it does not require labeling reagents and can be applied to an unlimited number of samples (Neilson et al., 2011; Evans et al., 2012). This is particularly appealing within plant proteomics, because the most conventional labeling strategy, Stable Isotope Labeling by Amino Acids in Cell Culture, is not easily suited for quantitative plant proteomic studies. The average labeling efficiency achieved using exogenous amino acid supply to Arabidopsis (Arabidopsis thaliana) cell cultures was found to be only 70% to 80% (Gruhler et al., 2005). Quantitative strategies with ¹⁵N metabolic labeling have been described for plant proteome analysis; however, care should be taken to ensure complete ¹⁵N incorporation, because even small amounts of ¹⁴N in the labeled sample can have significant detrimental effects on the number of peptide identifications (Nelson et al., 2007; Guo and Li, 2011; Arsova et al., 2012).In all label-free methods, samples under comparison are analyzed during separate mass spectrometry (MS) experiments (Neilson et al., 2011). The information from identified peptides is then used for relative and/or absolute quantification. The simplest label-free method involves taking the number of spectra acquired and assigned to peptides from the same protein as a measure of abundance (Ishihama et al., 2005). In an alternative approach, ion current recorded for a peptide ion is used as a measure of its abundance. The assumption is made that ion intensity is proportional to peptide amount in the sample analyzed, which holds true for nanoflow and microflow liquid chromatography (LC) systems (Levin et al., 2011; Christianson et al., 2013). Comparing peptide ion current between samples is, thus, widely used for relative quantification (Silva et al., 2005). To allow such comparison, a peptide must be identified across all samples under investigation, which is often challenging in LC-MS experiments given the highly complex nature of proteomics samples that contain tens of thousands of different peptides (Michalski et al., 2011). Hence, most relative ion intensity-based label-free approaches usually involve a step of identification transfer (Pasa-Tolíc et al., 2004). This involves matching ions from different acquisitions (in one of which, the ion has not been identified and is assigned the sequence from its matching pair in the other acquisition).Additionally, label-free proteomics can be used for absolute quantification (i.e. to estimate abundance of different proteins relative to each other within a given sample). Several different approaches have been suggested on how to convert peptide intensities to protein amounts (for comparison, see Wilhelm et al., 2014). One of the first such methods was Top-3 described by Silva et al. (2006b), who made a notable and unexpected observation, stating that the average MS signal response for the three most abundant peptides per 1 mol of protein is constant within a coefficient of variation of less than 10% (Silva et al., 2006b).In all these approaches, the peptide ion current is typically computed as the area under the curve of the chromatographic elution profile that is reconstituted from separate MS1 survey scans in which intact precursors are recorded. Determining a chromatographic profile accurately requires that the MS1 scans are performed at optimal frequency (Lange et al., 2008) and for optimal duration to record the MS1 signal at a high signal-to-noise ratio. In typical data-dependent acquisitions, however, the mass spectrometer oscillates between MS1 survey scans recording the mass/charge (m/z) for precursor peptide ions and then, a series of MS2 scans fragmenting one peptide ion precursor at a time, producing fragmentation spectra necessary for identification (Sadygov et al., 2004). As a result, the duration and frequency of MS2 scans determine the identification rate in data-dependent acquisition experiments but compromise time spent in MS1 required for accurate area under the curve quantification. Several groups have suggested data-independent acquisition, in which individual peptide ions are not selected for fragmentation but rather, groups of peptides of similar m/z are fragmented together. The exact number of cofragmented precursors depends on the speed and sensitivity of instrument configuration (for review, see Law and Lim, 2013). The simplest approach involves alternating between low-energy and high-energy scans of equal duration; low-energy scans record precursor peptide ions, whereas in high-energy scans, all precursors entering the mass spectrometer are cofragmented, and their fragments are recorded simultaneously. The method was called MS^E for Waters qTOF Mass Spectrometers (Geromanos et al., 2009) or all-ion fragmentation for Thermo Orbitrap Mass Spectrometers (Geiger et al., 2010). The analysis required downstream of this type of data acquisition is challenging given that the information of fragment origin (i.e. from what precursor peptide ion fragment was generated) is lost completely and that the high number of coeluting peptides is expected to create highly overlapping fragment spectra on fragmentation. To address this problem, Hoaglund-Hyzer and Clemmer (2001) have suggested fractionating peptides by ion mobility separation before fragmentation and MS and assigning fragments to precursors based on similarity of both chromatographic and mobility profiles (Hoaglund-Hyzer and Clemmer, 2001). The method was termed parallel fragmentation, and since that time, it has been commercialized by Waters as IMS-MS^E or HDMS^E (Shliaha et al., 2013).To date, the application of label-free quantitative proteomics to plant biology has been very limited. Recently, Helm et al. (2014) applied the LC-IMS-MS^E with Top-3 quantification to quantify the Arabidopsis chloroplast stroma proteome, allowing quantitative modeling of chloroplast metabolism. Two other works used the LC-MS^E method to assess the quantitative changes of cytosolic ribosomal proteins in response to Suc feeding and the extracellular proteome in response to salicylic acid (Cheng et al., 2009; Hummel et al., 2012).A number of proteomics approaches have been described to assess protein localization on a large scale (for review, see Gatto et al., 2010). Purification approaches attempt to isolate organelles to high levels of purity and subsequently identify and quantify proteins using LC-MS; however, such attempts yield limiting success and high false discovery rates (Andersen et al., 2002; Parsons et al., 2012a). A known limitation of this technique is the inability to completely isolate an organelle of interest, which combined with high proteome dynamic range, can result in some more abundant contaminants being identified and quantified at higher amounts than the target organelle residents. Moreover, even if a target organelle could be isolated to a certain degree of purity, it would still be impossible to deconvolute organelle residents from transient proteins that traffic through the target organelle. This becomes especially challenging for the organelles of the secretory pathway. To address these challenges, several groups applied fractionation of all organelles by gradient centrifugation and subsequent protein quantification by LC-MS. This produces distributions across the gradient for all quantified proteins, which are then used to assign organelle localization based on the specific distributions of organelle marker proteins. This effectively solves the problem of organelle contamination and protein trafficking, because a protein is expected to have a distribution characteristic of its organelle of residence, even if it is identified in all fractions, including those enriched in other organelles. Current variations of this method differ mostly by the LC-MS strategy used for quantification; for example, spectral counting was applied for protein-correlating profiles (Andersen et al., 2003), isobaric mass tagging (Nikolovski et al., 2012) and isotope-coded affinity tagging (Dunkley et al., 2004) were applied for localization of organelle proteins by isotope tagging (LOPIT), and Stable Isotope Labeling by Amino Acids in Cell Culture was applied for nucleolus/nucleus/cytosolic fractionation (Boisvert and Lamond, 2010).Here, a label-free LC-IMS-MS^E method was used for the analysis of density ultracentrifugation fractions enriched for the Golgi apparatus. First, we use relative label-free quantification involving identification transfer using the previously published synapter algorithm (Bond et al., 2013) to assess distributions of Golgi-localized proteins across the density gradient. These distributions are significantly different from those of residents of other organelles, which results in unambiguous protein assignment to the Golgi apparatus by multivariate data analysis. Second, the Top-3 absolute quantification method as implemented in Protein Lynx Global Server (PLGS) was used to rank order the Golgi-localized proteins by abundance in the fraction most enriched for Golgi apparatus. In conclusion, we present the analysis of protein distribution and abundances of the Golgi apparatus-enriched portion of the ultracentrifugation density gradient, allowing for simultaneous protein quantification and localization and leading to the assessment of relative abundances of 102 Golgi-localized proteins.