Short Abstract: The large diversity of microscopy applications requires sophisticated, individually adapted image processing methods that are able to process huge amounts of data, high background signals as well as low contrast images. The correct identification of objects or structures in micro- or nanometer range is crucial for diagnosing diseases, the determination of their progress and moreover for the impact of medical treatments, respectively.

For this case an image analysis suite has been developed for the automated fluorescent structure detection in various microscopy images containing biological or medical content. Generally, the image analysis suite is able to handle image preprocessing and bulk data analysis as well as quantification of single molecule localization and movement over a series of images. Moreover, it facilitates a very precise object detection of e.g., analysis of in-vivo protein-protein interactions in the cell membrane based on a fluorescence microscopy based approach (µ-patterning assay) or the identification of serotonin receptors in forebrain slices of depressive patients.
A strong background-heterogeneity of the brain slices sets high requirements on the preprocessing algorithms. In this particular case background reduction method such as á trous wavelets, noise filtering and intensity normalization are implemented for single molecule detection on localization microscopy images (STORM). In combination with an efficient single molecule fitting approach single molecules were automatically localized with position accuracy (PA) down to ~30 nm.

In future, machine learning algorithms from HeuristicLab (http://dev.heuristiclab.com) are exported and introduced in the image suite to produce more accurate and optimized object or tracking detection in microscopy images.

Short Abstract: PapC and FimD usher proteins are the outer membrane platforms responsible for the assembly and secretion of P and type-1 pili respectively. These adhesive appendages are responsible for host attachment and recognition in uropathogenic Escherichia coli (UPEC).
The usher comprises a transmembrane domain (TD) occluded by a plug domain (PD) (in the apo-form), and an N- and two C-terminal periplasmic domains. The TD comprises a 24 β-stranded barrel characterized by the presence of a conformationally-constrained β-hairpin capped by an α-helix. In the usher-chaperone-adhesin complex [1] the PD is displaced to accommodate the adhesin. Studies [2,3] suggest that this gating mechanism is likely controlled by two secondary structures within the TD, a β-hairpin and an α-helix.
To investigate the role of these structure elements in the allosteric signal propagation mechanism within the TD we performed molecular dynamics simulations of the native TD embedded in a POPE-POPG lipid bilayer and of 3 mutants in which the β-hairpin, the α-helix, and both, are deleted. By combining the conformational fluctuations information and the occupancy of the non-covalent interactions in the different systems with information derived from sequence evolution analysis, we derived a unique interaction network that connects the helix, the linkers of the PD and part of the barrel wall. This set of identified interactions can be potentially important for the propagation of the allosteric signal within the TD.

[1] Phan et al. Nature (2011), 474:49-53.
[2] Remaut et al. Cell (2008), 133(4):640-652.
[3] Mapingire et al. J Biol Chem (2009), 284(52):36324–36333

Short Abstract: We present an approach for the comprehensive integration of varied experimental data to study the spatial organization of human genome architectures. To address the challenge of modeling highly variable genome structures, we propose a population-based modeling approach, where we construct a large population of 3D genome structures that together are entirely consistent with all available experimental data, such as conformation capture experiments (TCC or Hi-C) and Fluorescent imaging experiments. We define a scoring function as the sum of all spatial constraints that act together on all models in the population. To create the population, we design an optimization protocol that evolves random initial configurations to a population of structures that completely satisfy all the constraints. We interpret the result in terms of probabilities of a sample drawn from a population of heterogeneous structures. Knowing the spatial probability distributions of individual genes provides deep insights into gene regulatory and replication processes, and fill in the missing links between functional genomics and structural biology.

Short Abstract: In this work, six different types of Particle Swarm Optimization algorithms (PSO) were tested with three distinct topologies (random, ring and global) for molecular docking of ten complexes, ranging from zero to twenty rotatable bonds. Each topology was tested with six distinct sets of parameters w (weight or inertia_factor), c1 (cognitive_factor) and c2 (social_factor). The six types of PSO were Constriction PSO (CPSO), Constant Weight PSO (CWPSO), Variable Weight PSO (VWPSO), Random pbest PSO (RBPSO), Fully-Informed PSO (FIPSO) and a hybrid algorithm consisting of Differential Evolution and PSO (DEPSO). The algorithms were embedded in the AutoDock3 and the results were compared with the Lamarkian Genetic Algorithm (LGA) of the AutoDock3 itself and with a previous PSO implementation by Vigneshwaran et al., embedded in the AutoDock3 as well. The variant FIPSO presented better RMSD, energy values and convergence than all other implementations, making it a valuable option for molecular docking, especially for highly flexible ligands such as peptides. In conclusion, our results show that, in general, PSO is a better optimization method than LGA for docking moderate to high flexible ligands. We also demonstrate that this result does not depend on the specific PSO variant but notably on the parameters used in it: the lower the parameters values used smaller is the contribution of the speed in the swarm motion and a better convergence, smaller energy and RMSD values are achieved. Currently, we are extending our test set to protein complexes containing ligands with 21 to 32 rotatable bonds.