This paper presents the concept detector module developed for the VITALAS multimedia retrieval system. It outlines its architecture and major implementation aspects, including a set of procedures and tools that were used for the development of detectors for more than 500 concepts. The focus is on aspects that increase the system's scalability in terms of the number of concepts: collaborative concept definition and disambiguation, selection of small but sufficient training sets and efficient manual annotation. The proposed architecture uses cross-domain concept fusion to improve effectiveness and reduce the number of samples required for concept detector training. Two criteria are proposed for selecting the best predictors to use for fusion and their effectiveness is experimentally evaluated for 221 concepts on the TRECVID-2005 development set and 132 concepts on a set of images provided by the Belga news agency. In these experiments, cross-domain concept fusion performed better than early fusion for most concepts. Experiments with variable training set sizes also indicate that cross-domain concept fusion is more effective than early fusion when the training set size is small.

MOTIVATIONS: Tissue engineering constitutes an important field with its potential of addressing the current shortage in organ availability. To successfully develop tissue-engineered organs, it is crucial to understand how to maintain the cells under conditions that maximize their ability to perform their physiological roles, regardless of the environment, whether the cells are part of an extracorporeal system, such as the bioartificial liver assist device, or an implantable tissue-engineered device. Our goals are to (1) provide insight into how cells will behave when confronted with changes in its environment and (2) determine the optimal environmental factors to achieve a desired level of cellular function. RESULTS: Diverse sets of environmental factors were used to systematically perturb the metabolic behavior associated with pre-conditioning and plasma supplementation. To probe metabolic state of hepatocytes, metabolic flux analysis was used to obtain the metabolic profile. We applied a multi-block partial least square (MPLS) model to relate environmental factors and fluxes to levels of intracellular lipids and urea synthesis. The MPLS model identified: (1) the most influential environmental factors and (2) how the metabolic pathways are altered by these factors. Finally, we inverted the MPLS model to determine the concentrations and types of environmental factors required to obtain the most economical solution for achieving optimal levels of cellular function for practical situations.

MOTIVATIONS: Classification of biological samples for diagnostic purposes is a difficult task because of the many decisions involved on the number, type and functional manipulations of the input variables. This study presents a generally applicable strategy for systematic formulation of optimal diagnostic indexes. To this end, we develop a novel set of computational tools by integrating regression optimization, stepwise variable selection and cross-validation algorithms. RESULTS: The proposed discrimination methodology was applied to plasma and tissue (liver) metabolic profiling data describing the time progression of liver dysfunction in a rat model of acute hepatic failure generated by d-galactosamine (GalN) injection. From the plasma data, our methodology identified seven (out of a total of 23) metabolites, and the corresponding transform functions, as the best inputs to the optimal diagnostic index. This index showed better time resolution and increased noise robustness compared with an existing metabolic index, Fischer's BCAA/AAA molar ratio, as well as indexes generated using other commonly used discriminant analysis tools. Comparison of plasma and liver indexes found two consensus metabolites, lactate and glucose, which implicate glycolysis and/or gluconeogenesis in mediating the metabolic effects of GalN.

MOTIVATION: Transcriptional profiling using microarrays can reveal important information about cellular and tissue expression phenotypes, but these measurements are costly and time consuming. Additionally, tissue sample availability poses further constraints on the number of arrays that can be analyzed in connection with a particular disease or state of interest. It is therefore important to provide a method for the determination of the minimum number of microarrays required to separate, with statistical reliability, distinct disease states or other physiological differences. RESULTS: Power analysis was applied to estimate the minimum sample size required for two-class and multi-class discrimination. The power analysis algorithm calculates the appropriate sample size for discrimination of phenotypic subtypes in a reduced dimensional space obtained by Fisher discriminant analysis (FDA). This approach was tested by applying the algorithm to existing data sets for estimation of the minimum sample size required for drawing certain conclusions on multi-class distinction with statistical reliability. It was confirmed that when the minimum number of samples estimated from power analysis is used, group means in the FDA discrimination space are statistically different. CONTACT: [email protected]

MOTIVATION: The increasing use of DNA microarrays to probe cell physiology requires methods for visualizing different expression phenotypes and explicitly connecting individual genes to discriminating expression features. Such methods should be robust and maintain biological interpretability. RESULTS: We propose a method for the mapping of the physiological state of cells and tissues from multidimensional expression data such as those obtained with DNA microarrays. The method uses Fisher discriminant analysis to create a linear projection of gene expression measurements that maximizes the separation of different sample classes. Relative to other typical classification methods, this method provides insights into the discriminating characteristics of expression measurements in terms of the contribution of individual genes to the definition of distinct physiological states. This projection method also facilitates visualization of classification results in a reduced dimensional space. Examples from four different cases demonstrate the ability of the method to produce well-separated groups in the projection space and to identify important genes for defining physiological states. The method can be augmented to also include data from the proteomic and metabolic phenotypes and can be useful in disease diagnosis, drug screening and bioprocessing applications.

The goal of this article is to present a methodology for the automatic acquisition of new control knowledge in the form of dominance and equivalence conditions, using an explanation-based learning algorithm from artificial intelligence. The acquisition of new control knowledge proceeds through the following stages: 1) Analysis of the problem-solving activity, generated from the application of a branch and bound algorithm on specific instance(s) of a class of problems. 2) Interpretation of the problem-solving activity and synthesis of new control knowledge, which can lead to more efficient solution of future problems. The generated control knowledge is provably correct within the theory of the given class of problems and the employed branch and bound strategy. Consequently, the number of nodes evaluated by a specific branch and bound algorithm is guaranteed to be reduced, as new control knowledge is continuously acquired from the solution of specific problems. The article presents the theoretical foundations for the handling of the above two tasks. It concludes with an application of the proposed approach to a mixed integer linear programming formulation of a batch scheduling problem.