Benchmarking protein classification algorithms via supervised cross-validation

Development and testing of protein classification algorithms are hampered by the fact that the protein universe is characterized by groups vastly different in the number of members, in average protein size, similarity within group, etc. Datasets based on traditional cross-validation (k-fold, leave-one-out, etc.) may not give reliable estimates on how an algorithm will generalize to novel, distantly related subtypes of the known protein classes. Supervised cross-validation, i.e., selection of test and train sets according to the known subtypes within a database has been successfully used earlier in conjunction with the SCOP database. Our goal was to extend this principle to other databases and to design standardized benchmark datasets for protein classification. Hierarchical classification trees of protein categories provide a simple and general framework for designing supervised cross-validation strategies for protein classification. Benchmark datasets can be designed at various levels of the concept hierarchy using a simple graph-theoretic distance. A combination of supervised and random sampling was selected to construct reduced size model datasets, suitable for algorithm comparison. Over 3000 new classification tasks were added to our recently established protein classification benchmark collection that currently includes protein sequence (including protein domains and entire proteins), protein structure and reading frame DNA sequence data. We carried out an extensive evaluation based on various machine-learning algorithms such as nearest neighbor, support vector machines, artificial neural networks, random forests and logistic regression, used in conjunction with comparison algorithms, BLAST, Smith-Waterman, Needleman-Wunsch, as well as 3D comparison methods DALI and PRIDE. The resulting datasets provide lower, and in our opinion more realistic estimates of the classifier performance than do random cross-validation schemes. A combination of supervised and random sampling was used to construct model datasets, suitable for algorithm comparison.

The datasets are available at http://hydra.icgeb.trieste.it/benchmark.     

Human cells and cell membrane molecular models are affected in vitro by chlorpromazine

This study presents evidence that chlorpromazine (CPZ) affects human cells and cell membrane molecular models. Human SH-SY5Y neuroblastoma cells incubated with 0.1 mM CPZ suffered a decrease of cell viability. On the other hand, phase contrast microscopy observations of human erythrocytes indicated that they underwent a morphological alteration as 1 μM CPZ changed their discoid normal shape to stomatocytes, and to hemolysis with 1 mM CPZ. X-ray diffraction experiments performed on dimyristoylphosphatidylcholine (DMPC) and dimyristoylphosphatidylethanolamine (DMPE) bilayers, classes of the major phospholipids present in the outer and inner sides of the erythrocyte membrane, respectively showed that CPZ disordered the polar head and acyl chain regions of both DMPC and DMPE, where these interactions were stronger with DMPC bilayers. Fluorescence spectroscopy on DMPC LUV at 18 °C confirmed these results. In fact, the assays showed that CPZ induced a significant reduction of their generalized polarization (GP) and anisotropy (r) values, indicative of enhanced disorder at the polar head and acyl chain regions of the DMPC lipid bilayer.     

Radial glia and neural stem cells

During the last decade, the role of radial glia has been radically revisited. Rather than being considered a mere structural component serving to guide newborn neurons towards their final destinations, radial glia is now known to be the main source of neurons in several regions of the central nervous system, notably in the cerebral cortex. Radial glial cells differentiate from neuroepithelial progenitors at the beginning of neurogenesis and share with their ancestors the bipolar shape and the expression of some molecular markers. Radial glia, however, can be distinguished from neuroepithelial progenitors by the expression of astroglial markers. Clonal analyses showed that radial glia is a heterogeneous population, comprising both pluripotent and different lineage-restricted neural progenitors. At late-embryonic and postnatal stages, radial glial cells give rise to the neural stem cells responsible for adult neurogenesis. Embryonic pluripotent radial glia and adult neural stem cells may be clonally linked, thus representing a lineage displaying stem cell features in both the developing and mature central nervous system. This work was supported by AIRC (Associazione Italiana per la Ricerca sul Cancro) NUSUG grant (In vivo screening for genes implicated in glioma formation and development of new animal models of glial tumors) and by Fondazione CARIGE grant (Basi molecolari e cellulari dei gliomi: individuazione di marcatori diagnostici e di nuovi bersagli terapeutici).     

Human fetal aorta-derived vascular progenitor cells: identification and potential application in ischemic diseases

Vasculogenesis, the formation of blood vessels in embryonic or fetal tissue mediated by immature vascular cells (i.e., angioblasts), is poorly understood. Here we report a summary of our recent studies on the identification of a population of vascular progenitor cells (VPCs) in human fetal aorta. These undifferentiated mesenchymal cells co-express endothelial and myogenic markers (CD133+, CD34+, KDR+, desmin+) and are localized in outer layer of the aortic stroma of 11–12 weeks old human fetuses. Under stimulation with VEGF-A or PDGF-BB, VPCs give origin to a mixed population of mature endothelial and mural cells, respectively. When embedded in a three-dimensional collagen gel, VPCs organize into cohesive cellular cords that resembled mature vascular structures. The therapeutic efficacy of a small number of VPCs transplanted into ischemic limb muscle was demonstrated in immunodeficient mice. Investigation of the effect of VPCs on experimental heart ischemia and on diabetic ischemic ulcers in mice is in progress and seems to confirm their efficacy. On the whole, fetal aorta represents an important source for the investigation of phenotypic and functional features of human vascular progenitor cells.     

Characterization of lactic acid bacteria isolated from sourdoughs for Cornetto, a traditional bread produced in Basilicata (Southern Italy)

A total of 41 strains of lactic acid bacteria (LAB) isolated from durum wheat sourdoughs used to produce Cornetto di Matera bread, were identified by SDS-PAGE of whole cell proteins (WCP) and screened for acid production ability, antimicrobial activity and exopolysaccharide (EPS) production. The isolates were identified as Lactobacillus plantarum (49%), Leuconostoc mesenteroides (17%), Lactobacillus curvatus (15%), Lactobacillus paraplantarum (12%), Weissella cibaria (5%) and Lactobacillus pentosus (2%). Several strains of Lb. plantarum and Leuc. mesenteroides showed a high acid production ability. The antagonistic activity was tested using an agar-spot deferred antagonism assay against a set of five indicators. The species had different profiles of inhibition. Lb. plantarum had the largest spectrum of inhibition, while no isolates of W. cibaria and Leuc. mesenteroides showed antimicrobial activity. No strains had antimicrobial activity against Bacillus cereus. The inhibitory activity of five strains was confirmed to be sensitive to proteolytic enzymes and thus potentially due to bacteriocin production. All Leuc. mesenteroides and W. cibaria strains produced EPS from sucrose. Some Lb. plantarum and Lb. paraplantarum strains produced EPS from different sugars in solid media. EPS production in liquid media was different within the species, with the highest production in liquid media containing glucose and maltose. A defined strain starter culture (W. cibaria DBPZ1006, Lb. plantarum DBPZ1015 and S. cerevisiae MTG10) was selected on the basis of technological properties and tested in model sourdough fermentations.     

Microseconds dynamics simulations of the outer-membrane protease T

Conformational fluctuations of enzymes may play an important role for substrate recognition and/or catalysis, as it has been suggested in the case of the protease enzymatic superfamily. Unfortunately, theoretically addressing this issue is a problem of formidable complexity, as the number of the involved degrees of freedom is enormous: indeed, the biological function of a protein depends, in principle, on all its atoms and on the surrounding water molecules. Here we investigated a membrane protease enzyme, the OmpT from Escherichia coli, by a hybrid molecular mechanics/coarse-grained approach, in which the active site is treated with the GROMOS force field, whereas the protein scaffold is described with a Go-model. The method has been previously tested against results obtained with all-atom simulations. Our results show that the large-scale motions and fluctuations of the electric field in the microsecond timescale may impact on the biological function and suggest that OmpT employs the same catalytic strategy as aspartic proteases. Such a conclusion cannot be drawn within the 10- to 100-ns timescale typical of current molecular dynamics simulations. In addition, our studies provide a structural explanation for the drop in the catalytic activity of two known mutants (S99A and H212A), suggesting that the coarse-grained approach is a fast and reliable tool for providing structure/function relationships for both wild-type OmpT and mutants.     

Contact guidance mediated three-dimensional cell migration is regulated by Rho/ROCK-dependent matrix reorganization

Cells generate mechanical force to organize the extracellular matrix (ECM) and drive important developmental and reparative processes. Likewise, tumor cells invading into three-dimensional (3D) matrices remodel the ECM microenvironment. Importantly, we previously reported a distinct radial reorganization of the collagen matrix surrounding tumors that facilitates local invasion. Here we describe a mechanism by which cells utilize contractility events to reorganize the ECM to provide contact guidance that facilitates 3D migration. Using novel assays to differentially organize the collagen matrix we show that alignment of collagen perpendicular to the tumor-explant boundary promotes local invasion of both human and mouse mammary epithelial cells. In contrast, organizing the collagen matrix to mimic the ECM organization associated with noninvading regions of tumors suppresses 3D migration/invasion. Moreover, we demonstrate that matrix reorganization is contractility-dependent and that the Rho/Rho kinase pathway is necessary for collagen alignment to provide contact guidance. Yet, if matrices are prealigned, inhibiting neither Rho nor Rho kinase inhibits 3D migration, which supports our conclusion that Rho-mediated matrix alignment is an early step in the invasion process, preceding and subsequently facilitating 3D migration.     

Diffusion of the second messengers in the cytoplasm acts as a variability suppressor of the single photon response in vertebrate phototransduction

The single photon response in vertebrate phototransduction is highly reproducible despite a number of random components of the activation cascade, including the random activation site, the random walk of an activated receptor, and its quenching in a random number of steps. Here we use a previously generated and tested spatiotemporal mathematical and computational model to identify possible mechanisms of variability reduction. The model permits one to separate the process into modules, and to analyze their impact separately. We show that the activation cascade is responsible for generation of variability, whereas diffusion of the second messengers is responsible for its suppression. Randomness of the activation site contributes at early times to the coefficient of variation of the photoresponse, whereas the Brownian path of a photoisomerized rhodopsin (Rh*) has a negligible effect. The major driver of variability is the turnoff mechanism of Rh*, which occurs essentially within the first 2-4 phosphorylated states of Rh*. Theoretically increasing the number of steps to quenching does not significantly decrease the corresponding coefficient of variation of the effector, in agreement with the biochemical limitations on the phosphorylated states of the receptor. Diffusion of the second messengers in the cytosol acts as a suppressor of the variability generated by the activation cascade. Calcium feedback has a negligible regulatory effect on the photocurrent variability. A comparative variability analysis has been conducted for the phototransduction in mouse and salamander, including a study of the effects of their anatomical differences such as incisures and photoreceptors geometry on variability generation and suppression.