Cluster-based active learning for compact image classification

In this paper, we consider active sampling to label pixels grouped with hierarchical clustering. The objective of the method is to match the data relationships discovered by the clustering algorithm with the user's desired class semantics. The first is represented as a complete tree to be pruned and the second is iteratively provided by the user. The active learning algorithm proposed searches the pruning of the tree that best matches the labels of the sampled points. By choosing the part of the tree to sample from according to current pruning's uncertainty, sampling is focused on most uncertain clusters. This way, large clusters for which the class membership is already fixed are no longer queried and sampling is focused on division of clusters showing mixed labels. The model is tested on a VHR image in a multiclass classification setting. The method clearly outperforms random sampling in a transductive setting, but cannot generalize to unseen data, since it aims at optimizing the classification of a given cluster structure.

β1- and β3- voltage-gated sodium channel subunits modulate cell surface expression and glycosylation of Nav1.7 in HEK293 cells.

Voltage-gated sodium channels (Navs) are glycoproteins composed of a pore-forming α-subunit and associated β-subunits that regulate Nav α-subunit plasma membrane density and biophysical properties. Glycosylation of the Nav α-subunit also directly affects Navs gating. β-subunits and glycosylation thus comodulate Nav α-subunit gating. We hypothesized that β-subunits could directly influence α-subunit glycosylation. Whole-cell patch clamp of HEK293 cells revealed that both β1- and β3-subunits coexpression shifted V ½ of steady-state activation and inactivation and increased Nav1.7-mediated I Na density. Biotinylation of cell surface proteins, combined with the use of deglycosydases, confirmed that Nav1.7 α-subunits exist in multiple glycosylated states. The α-subunit intracellular fraction was found in a core-glycosylated state, migrating at ~250 kDa. At the plasma membrane, in addition to the core-glycosylated form, a fully glycosylated form of Nav1.7 (~280 kDa) was observed. This higher band shifted to an intermediate band (~260 kDa) when β1-subunits were coexpressed, suggesting that the β1-subunit promotes an alternative glycosylated form of Nav1.7. Furthermore, the β1-subunit increased the expression of this alternative glycosylated form and the β3-subunit increased the expression of the core-glycosylated form of Nav1.7. This study describes a novel role for β1- and β3-subunits in the modulation of Nav1.7 α-subunit glycosylation and cell surface expression.

Recruitment of Matrix Metalloproteinase-9 (MMP-9) to the Fibroblast Cell Surface by Lysyl Hydroxylase 3 (LH3) Triggers Transforming Growth Factor-β (TGF-β) Activation and Fibroblast Differentiation.

Solid tumor growth triggers a wound healing response. Similar to wound healing, fibroblasts in the tumor stroma differentiate into myofibroblasts (also referred to as cancer-associated fibroblasts) primarily, but not exclusively, in response to transforming growth factor-β (TGF-β). Myofibroblasts in turn enhance tumor progression by remodeling the stroma. Among proteases implicated in stroma remodeling, matrix metalloproteinases (MMPs), including MMP-9, play a prominent role. Recent evidence indicates that MMP-9 recruitment to the tumor cell surface enhances tumor growth and invasion. In the present work, we addressed the potential relevance of MMP-9 recruitment to and activity at the surface of fibroblasts. We show that recruitment of MMP-9 to the fibroblast cell surface occurs through its fibronectin-like (FN) domain and that the molecule responsible for the recruitment is lysyl hydroxylase 3 (LH3). Functional assays suggest that both pro- and active MMP-9 trigger α-smooth muscle actin expression in cultured fibroblasts, reflecting myofibroblast differentiation, possibly as a result of TGF-β activation. Moreover, the recombinant FN domain inhibited both MMP-9-induced TGF-β activation and α-smooth muscle actin expression by displacing MMP-9 from the fibroblast cell surface. Together our results uncover LH3 as a new docking receptor of MMP-9 on the fibroblast cell surface and demonstrate that the MMP-9 FN domain is essential for the interaction. They also show that the recombinant FN domain inhibits MMP-9-induced TGF-β activation and fibroblast differentiation, providing a potentially attractive therapeutic reagent toward attenuating tumor progression where MMP-9 activity is strongly implicated.

The Human Acid-Sensing Ion Channel ASIC1a: Evidence for a Homotetrameric Assembly State at the Cell Surface.

The chicken acid-sensing ion channel ASIC1 has been crystallized as a homotrimer. We address here the oligomeric state of the functional ASIC1 in situ at the cell surface. The oligomeric states of functional ASIC1a and mutants with additional cysteines introduced in the extracellular pore vestibule were resolved on SDS-PAGE. The functional ASIC1 complexes were stabilized at the cell surface of Xenopus laevis oocytes or CHO cells either using the sulfhydryl crosslinker BMOE, or sodium tetrathionate (NaTT). Under these different crosslinking conditions ASIC1a migrates as four distinct oligomeric states that correspond by mass to multiples of a single ASIC1a subunit. The relative importance of each of the four ASIC1a oligomers was critically dependent on the availability of cysteines in the transmembrane domain for crosslinking, consistent with the presence of ASIC1a homo-oligomers. The expression of ASIC1a monomers, trimeric or tetrameric concatemeric cDNA constructs resulted in functional channels. The resulting ASIC1a complexes are resolved as a predominant tetramer over the other oligomeric forms, after stabilization with BMOE or NaTT and SDS-PAGE/western blot analysis. Our data identify a major ASIC1a homotetramer at the surface membrane of the cell expressing functional ASIC1a channel.

From Regional Landslide Detection to Site-Specific Slope Deformation Monitoring and Modelling Based on Active Remote Sensors

Landslide processes can have direct and indirect consequences affecting human lives and activities. In order to improve landslide risk management procedures, this PhD thesis aims to investigate capabilities of active LiDAR and RaDAR sensors for landslides detection and characterization at regional scales, spatial risk assessment over large areas and slope instabilities monitoring and modelling at site-specific scales. At regional scales, we first demonstrated recent boat-based mobile LiDAR capabilities to model topography of the Normand coastal cliffs. By comparing annual acquisitions, we validated as well our approach to detect surface changes and thus map rock collapses, landslides and toe erosions affecting the shoreline at a county scale. Then, we applied a spaceborne InSAR approach to detect large slope instabilities in Argentina. Based on both phase and amplitude RaDAR signals, we extracted decisive information to detect, characterize and monitor two unknown extremely slow landslides, and to quantify water level variations of an involved close dam reservoir. Finally, advanced investigations on fragmental rockfall risk assessment were conducted along roads of the Val de Bagnes, by improving approaches of the Slope Angle Distribution and the FlowR software. Therefore, both rock-mass-failure susceptibilities and relative frequencies of block propagations were assessed and rockfall hazard and risk maps could be established at the valley scale. At slope-specific scales, in the Swiss Alps, we first integrated ground-based InSAR and terrestrial LiDAR acquisitions to map, monitor and model the Perraire rock slope deformation. By interpreting both methods individually and originally integrated as well, we therefore delimited the rockslide borders, computed volumes and highlighted non-uniform translational displacements along a wedge failure surface. Finally, we studied specific requirements and practical issues experimented on early warning systems of some of the most studied landslides worldwide. As a result, we highlighted valuable key recommendations to design new reliable systems; in addition, we also underlined conceptual issues that must be solved to improve current procedures. To sum up, the diversity of experimented situations brought an extensive experience that revealed the potential and limitations of both methods and highlighted as well the necessity of their complementary and integrated uses.