Parallel matrix factorization for recommender systems

Matrix factorization, when the matrix has missing values, has become one of the leading techniques for recommender systems. To handle web-scale datasets with millions of users and billions of ratings, scalability becomes an important issue. Alternating least squares (ALS) and stochastic gradient descent (SGD) are two popular approaches to compute matrix factorization, and there has been a recent flurry of activity to parallelize these algorithms. However, due to the cubic time complexity in the target rank, ALS is not scalable to large-scale datasets. On the other hand, SGD conducts efficient updates but usually suffers from slow convergence that is sensitive to the parameters. Coordinate descent, a classical optimization approach, has been used for many other large-scale problems, but its application to matrix factorization for recommender systems has not been thoroughly explored. In this paper, we show that coordinate descent-based methods have a more efficient update rule compared to ALS and have faster and more stable convergence than SGD. We study different update sequences and propose the CCD++ algorithm, which updates rank-one factors one by one. In addition, CCD++ can be easily parallelized on both multi-core and distributed systems. We empirically show that CCD++ is much faster than ALS and SGD in both settings. As an example, with a synthetic dataset containing 14.6 billion ratings, on a distributed memory cluster with 64 processors, to deliver the desired test RMSE, CCD++ is 49 times faster than SGD and 20 times faster than ALS. When the number of processors is increased to 256, CCD++ takes only 16 s and is still 40 times faster than SGD and 20 times faster than ALS.     

Identification of the pathway of α-oxidation of cerebronic acid in peroxisomes

Cerebronic acid (2-hydroxytetracosanoic acid), an α-hydroxy very long-chain fatty acid (VLCFA) and a component of cerebrosides and sulfatides, is unique to nervous tissues. Studies were carried out to identify the pathway and the subcellular site involved in the oxidation of cerebronic acid. The results from these studies revealed that cerebronic acid was catabolized by α-oxidation to CO₂ and tricosanoic acid (23:0). Studies with subcellular fractions indicated that cerebronic acid was α-oxidized in fractions having particulate bound catalase and enzyme systems for the β-oxidation of VLCFA (e.g., lignoceric acid), suggesting peroxisomes as the subcellular organelle responsible for α-oxidation of cerebronic acid. Etomoxir, an inhibitor of mitochondrial fatty acid oxidation, had no effect on cerebronic acid α-oxidation. Further, cerebronic acid oxidation was found to be dependent on the presence of NAD⁺ but not FAD, NADPH, ATP, Mg²⁺, or CoASH. Intraorganellar localization studies indicated that the enzyme system for the α-oxidation of cerebronic acid was associated with the peroxisomal limiting membranes. Studies on cultured fibroblasts from normal subjects and patients with peroxisomal disorders indicated an impairment of α-oxidation of cerebronic acid in cell lines that lack peroxisomes [e.g., Zellweger syndrome (ZS)]. On the other hand, α-oxidation of cerebronic acid was found to be normal in cell lines from X-linked adrenoleukodystrophy, adult Refsum disease, and rhizomelic chondrodysplasia punctata. Our results clearly demonstrate that α-oxidation of α-hydroxy VLCFA (cerebronic acid) is a peroxisomal function and that this oxidation is impaired in ZS. Furthermore, this α-oxidation enzyme system is distinct from the one for the α-oxidation of β-carbon branched-chain fatty acids (e.g., phytanic acid).     

Are Laboratory Bioassays for Allelopathy Suitable for Prediction of Field Responses?

One concern often voiced by researchers of allelopathic interactions is that many laboratory bioassays do not adequately predict the responses observed in field situations. The questions that arise are: (1) What criteria should be implemented to design ecologically relevant bioassays? (2) What species (crops or weeds) are involved in the interaction? (3) Are we investigating allelopathy of debris/residues or interactions involving living plants? (4) Which plant indicator species are actually cohabiting with the species under investigation? and (5) What are appropriate experimental controls? It is difficult to design a bioassay that can be used to examine responses in all species. In fact, each bioassay must be designed specifically to assess species interactions after careful consideration of growth habit, biotic characteristics, and ecophysiological factors. The objective of this paper is to discuss the significance of bioassays designed to study a particular aspect of allelopathy. We conclude that through a laboratory bioassay we can not demonstrate that allelopathy is operational in natural settings. An investigator should consider allelopathy as one component in a multifaceted approach to ecology and address key questions to determine the relevance of a particular assay. Due to the complexity of field interactions and responses, one can only hope to predict and describe some of the cause-and-effect relationships observed in a field setting. An accurate assessment of these main effects will prove invaluable in directing the focus of future research emphasis.     

Protein modeling and active site binding mode interactions of myrosinase-sinigrin in Brassica juncea--an in silico approach

Myrosinase, the only known S-glycosidase, occurs particularly in Cruciferae family. It is responsible for the hydrolysis of glucosinolates and serves as a vital element of plant defense system. The biological and chemical properties of myrosinase catalyzed products of glucosinolates are well characterized. The myrosinase-protein-sequence of Brassica juncea was retrieved from NCBI database and its 3-D model was generated on the basis of crystal structure of 1MYR-A, 1E4M-M and 1DWA-M chains of myrosinase from Sinapis alba by employing Modeller9v7 program. Homolog templates from S. alba exhibited 72% identity with target sequence. The model was optimized by using molecular dynamics (MD) approach together with simulated annealing (SA) methods in the same Modeller program, and eventually verified and validated on SAVES (Structure Analysis and Verification Server) and PROCHECK programs, respectively. Ramachandran plot obtained through PROCHECK program depicted that 99.8% of total residues were confined to the allowed region while only one residue (Thr92) was restrained to the disallowed region. Additionally, B. juncea myrosinase contains three disulphide bridges which were found to be conserved in S. alba homologs as well. Further, overlapping of B. juncea myrosinase with that of template protein 1MYR-A from S. alba stipulates the amino acid residues Arg115, Gln207, Thr210, Asn350, Tyr352 and Glu429 that constitute active site of the enzyme. Active site analysis also speculates the presence of a hydrophobic pocket in addition to seven N-glycosylation sites. Docking studies of enzyme and substrate illuminate the interactions of various active site residues with diverse groups of sinigrin. Therefore, the present study furnishes the first significant, in silico insight into the 3-D structure, active site machinery, and enzyme-substrate interactions of B. juncea myrosinase.     

Phytotoxicity and fate of 1,1,2-trichloroethylene: a laboratory study

1,1,2-Trichloroethylene (TCE), a chlorinated organic contaminant, poses serious environmental concerns. A study was conducted to evaluate the phytotoxicity of TCE to a crop species and its fate in vermiculite. Growth bioassays were carried out using carrot (Daucus carota L.) as the test species. Three different concentrations, 0.25, 0.50, and 1 ppm were used to evaluate phytotoxicity of TCE. When added to petri plates with cotton pads, TCE did not have any effect on carrot seedling growth. However, when added to vermiculite, it significantly suppressed growth. Shoot growth was inhibited only at the 1 ppm concentration. Recovery experiments were carried out to study the fate of TCE in vermiculite. A significant decline in the percent recovery was observed with time. Interestingly, TCE additional peaks (unknown organic molecules) were detected with declining concentrations. The available chloride ion concentration in vermiculite containing 1 ppm of TCE for 24 hr was significantly higher compared to control.     

Growth and physiological responses of Black Spruce (Picea mariana) to sites dominated byLedum groenlandicum

Growth and physiological parameters of black spruce planted eight years previously were compared at sites with and withoutLedum groenlandicum to quantify the adverse effects on black spruce inLedum sites. Tree height, crown diameter, and dry weights of stems, branches, and needles of black spruce were significantly less at theLedum-dominated sites. Annual stem height, basal diameter, growth ring width, and cumulative wood volume of black spruce were significantly reduced in theLedum sites compared to those of the non-Ledum sites. Yearly growth of black spruce in the first seven years after planting was more affected in theLedum site than the eighth year's growth. Foliar nutrient analysis of black spruce in theLedum site showed an overall imbalance in nutrient accumulation. Black spruce growing in theLedum site had higher foliar concentrations of Al, Ca, Fe, Ni, and S and lower concentrations of N and Si than those growing in the non-Ledum site. Black spruce growing in theLedum site also had a higher foliar phenolic content. Preliminary experiments suggest thatLedum has the potential to leach water-soluble phenolics to its immediate organic soil.     

Free Convective Mass Transfer Behaviour of Inclined Cylinders with Active Ends

Rates of mass transfer at inclined cylinders with active ends were determined under natural convection conditions by measuring the limiting current of copper deposition from acidified copper sulphate. Physical properties of the solution, cylinder dimensions and inclination were varied to produce a ScGr range of 1.53 × 10⁹ to 7 × 10¹⁰. Under these conditions the data for different orientations were correlated by Sh = 1.075 (ScGr)^0.242. The characteristic length used in obtaining Sh and Gr was calculated by dividing the cylinder surface area by the cylinder perimeter projected onto the horizontal plane. At high CuSO₄ concentrations a reasonable agreement was found between the experimental rate of mass transfer and that calculated by summing the rate of mass transfer at individual surfaces of the cylinder. Deviations were observed at low CuSO₄ concentrations, and the possible causes of the deviations were explained.     

Formononetin 7-O-glucoside (ononin), an additional growth inhibitor in soils associated with the weed,Pluchea lanceolata (DC) C.B. Clarke (Asteraceae)

Formononetin 7-O-glucoside (ononin), an isoflavonoid, was isolated from the soils of cultivated areas associated with the perennial weed,Pluchea lanceolata. Aqueous solutions of this compound inhibited significantly root and shoot growth of mustard at 1×10^–4 M, 5×10^–4 M, and 1×10^–3M. The level of inhibition was similar to that of hesperidin and taxifolin 3-arabinoside, as reported earlier. The potential allelopathic effect of this compound under field conditions is discussed.     

Rey-zeolite and silica-alumina mixed support for hydrotreating heavy gas oil

Rare earth exchanged Y-type zeolite (REY-zeolite) was dispersed in a silica-alumina gel to prepare catalyst supports with better hydrogenolysis activity. Such support material showed improved hydrotreating properties compared to commercial catalysts, especially for heavy gas oils. Statistical experimental designs used to optimize the composition of such mixed supports suggested a composition of 10 wt.% silica, 25 wt.% zeolite and 65 wt.% alumina as optimum for hydrotreating a heavy gas oil (343°C to 525°C fraction) obtained from hydrocracking of Athabasca bitumen. The kinetic parameters were then evolved for the optimum catalyst.     

Mass and heat transfer behavior of oscillating helical coils in relation to heterogeneous reactor design

Rates of mass and heat transfer at vibrating helical coils were studied by the electrochemical technique with the object of using helical coils as heat exchanger/reactor for conducting liquid–solid diffusion controlled reactions. Variables studied were frequency and amplitude of vibration, tube diameter, and superimposed axial flow velocity. The data for vibrating coil (batch reactor) were correlated for 59 < < 4965; Sc = 2314 by the equation: The data were found to be consistent with the analogy model. For vibrated helical coils with superimposed axial flow, the data were correlated by the equation: Importance of the present results in the design and operation of heterogeneous reactors used to conduct diffusion controlled exothermic reactions involving heat sensitive materials was pointed out. Also the importance of the present results in the design and operation of shell and helical tube heat exchangers of improved performance was highlighted. © 2017 American Institute of Chemical Engineers AIChE J, 63: 3141–3149, 2017