New definitions of margin for multi-class Boosting algorithms

La familia de algoritmos de Boosting son un tipo de técnicas de clasificación y regresión que han demostrado ser muy eficaces en problemas de Visión Computacional. Tal es el caso de los problemas de detección, de seguimiento o bien de reconocimiento de caras, personas, objetos deformables y acciones. El primer y más popular algoritmo de Boosting, AdaBoost, fue concebido para problemas binarios. Desde entonces, muchas han sido las propuestas que han aparecido con objeto de trasladarlo a otros dominios más generales: multiclase, multilabel, con costes, etc. Nuestro interés se centra en extender AdaBoost al terreno de la clasificación multiclase, considerándolo como un primer paso para posteriores ampliaciones. En la presente tesis proponemos dos algoritmos de Boosting para problemas multiclase basados en nuevas derivaciones del concepto margen. El primero de ellos, PIBoost, está concebido para abordar el problema descomponiéndolo en subproblemas binarios. Por un lado, usamos una codificación vectorial para representar etiquetas y, por otro, utilizamos la función de pérdida exponencial multiclase para evaluar las respuestas. Esta codificación produce un conjunto de valores margen que conllevan un rango de penalizaciones en caso de fallo y recompensas en caso de acierto. La optimización iterativa del modelo genera un proceso de Boosting asimétrico cuyos costes dependen del número de etiquetas separadas por cada clasificador débil. De este modo nuestro algoritmo de Boosting tiene en cuenta el desbalanceo debido a las clases a la hora de construir el clasificador. El resultado es un método bien fundamentado que extiende de manera canónica al AdaBoost original. El segundo algoritmo propuesto, BAdaCost, está concebido para problemas multiclase dotados de una matriz de costes. Motivados por los escasos trabajos dedicados a generalizar AdaBoost al terreno multiclase con costes, hemos propuesto un nuevo concepto de margen que, a su vez, permite derivar una función de pérdida adecuada para evaluar costes. Consideramos nuestro algoritmo como la extensión más canónica de AdaBoost para este tipo de problemas, ya que generaliza a los algoritmos SAMME, Cost-Sensitive AdaBoost y PIBoost. Por otro lado, sugerimos un simple procedimiento para calcular matrices de coste adecuadas para mejorar el rendimiento de Boosting a la hora de abordar problemas estándar y problemas con datos desbalanceados. Una serie de experimentos nos sirven para demostrar la efectividad de ambos métodos frente a otros conocidos algoritmos de Boosting multiclase en sus respectivas áreas. En dichos experimentos se usan bases de datos de referencia en el área de Machine Learning, en primer lugar para minimizar errores y en segundo lugar para minimizar costes. Además, hemos podido aplicar BAdaCost con éxito a un proceso de segmentación, un caso particular de problema con datos desbalanceados. Concluimos justificando el horizonte de futuro que encierra el marco de trabajo que presentamos, tanto por su aplicabilidad como por su flexibilidad teórica. Abstract The family of Boosting algorithms represents a type of classification and regression approach that has shown to be very effective in Computer Vision problems. Such is the case of detection, tracking and recognition of faces, people, deformable objects and actions. The first and most popular algorithm, AdaBoost, was introduced in the context of binary classification. Since then, many works have been proposed to extend it to the more general multi-class, multi-label, costsensitive, etc... domains. Our interest is centered in extending AdaBoost to two problems in the multi-class field, considering it a first step for upcoming generalizations. In this dissertation we propose two Boosting algorithms for multi-class classification based on new generalizations of the concept of margin. The first of them, PIBoost, is conceived to tackle the multi-class problem by solving many binary sub-problems. We use a vectorial codification to represent class labels and a multi-class exponential loss function to evaluate classifier responses. This representation produces a set of margin values that provide a range of penalties for failures and rewards for successes. The stagewise optimization of this model introduces an asymmetric Boosting procedure whose costs depend on the number of classes separated by each weak-learner. In this way the Boosting procedure takes into account class imbalances when building the ensemble. The resulting algorithm is a well grounded method that canonically extends the original AdaBoost. The second algorithm proposed, BAdaCost, is conceived for multi-class problems endowed with a cost matrix. Motivated by the few cost-sensitive extensions of AdaBoost to the multi-class field, we propose a new margin that, in turn, yields a new loss function appropriate for evaluating costs. Since BAdaCost generalizes SAMME, Cost-Sensitive AdaBoost and PIBoost algorithms, we consider our algorithm as a canonical extension of AdaBoost to this kind of problems. We additionally suggest a simple procedure to compute cost matrices that improve the performance of Boosting in standard and unbalanced problems. A set of experiments is carried out to demonstrate the effectiveness of both methods against other relevant Boosting algorithms in their respective areas. In the experiments we resort to benchmark data sets used in the Machine Learning community, firstly for minimizing classification errors and secondly for minimizing costs. In addition, we successfully applied BAdaCost to a segmentation task, a particular problem in presence of imbalanced data. We conclude the thesis justifying the horizon of future improvements encompassed in our framework, due to its applicability and theoretical flexibility.

Regulation of transport pathways in tumor vessels: Role of tumor type and microenvironment

Novel anti-neoplastic agents such as gene targeting vectors and encapsulated carriers are quite large (approximately 100–300 nm in diameter). An understanding of the functional size and physiological regulation of transvascular pathways is necessary to optimize delivery of these agents. Here we analyze the functional limits of transvascular transport and its modulation by the microenvironment. One human and five murine tumors including mammary and colorectal carcinomas, hepatoma, glioma, and sarcoma were implanted in the dorsal skin-fold chamber or cranial window, and the pore cutoff size, a functional measure of transvascular gap size, was determined. The microenvironment was modulated: (i) spatially, by growing tumors in subcutaneous or cranial locations and (ii) temporally, by inducing vascular regression in hormone-dependent tumors. Tumors grown subcutaneously exhibited a characteristic pore cutoff size ranging from 200 nm to 1.2 μm. This pore cutoff size was reduced in tumors grown in the cranium or in regressing tumors after hormone withdrawal. Vessels induced in basic fibroblast growth factor-containing gels had a pore cutoff size of 200 nm. Albumin permeability was independent of pore cutoff size. These results have three major implications for the delivery of therapeutic agents: (i) delivery may be less efficient in cranial tumors than in subcutaneous tumors, (ii) delivery may be reduced during tumor regression induced by hormonal ablation, and (iii) permeability to a molecule is independent of pore cutoff size as long as the diameter of the molecule is much less than the pore diameter.

Inhibition of protease-resistant prion protein formation by porphyrins and phthalocyanines

A central aspect of pathogenesis in the transmissible spongiform encephalopathies or prion diseases is the conversion of normal protease-sensitive prion protein (PrP-sen) to the abnormal protease-resistant form, PrP-res. Here we identify porphyrins and phthalocyanines as inhibitors of PrP-res accumulation. The most potent of these tetrapyrroles had IC50 values of 0.5–1 μM in scrapie-infected mouse neuroblastoma (ScNB) cell cultures. Inhibition was observed without effects on protein biosynthesis in general or PrP-sen biosynthesis in particular. Tetrapyrroles also inhibited PrP-res formation in a cell-free reaction composed predominantly of hamster PrP-res and PrP-sen. Inhibitors were found among phthalocyanines, deuteroporphyrins IX, and meso-substituted porphines; examples included compounds containing anionic, neutral protic, and cationic peripheral substituents and various metals. We conclude that certain tetrapyrroles specifically inhibit the conversion of PrP-sen to PrP-res without apparent cytotoxic effects. The inhibition observed in the cell-free conversion reaction suggests that the mechanism involved direct interactions of the tetrapyrrole with PrP-res and/or PrP-sen. These findings introduce a new class of inhibitors of PrP-res formation that represents a potential source of therapeutic agents for transmissible spongiform encephalopathies.

Immunotargeting of liposomes to activated vascular endothelial cells: A strategy for site-selective delivery in the cardiovascular system

Endothelial-selective delivery of therapeutic agents, such as drugs or genes, would provide a useful tool for modifying vascular function in various disease states. A potential molecular target for such delivery is E-selectin, an endothelial-specific cell surface molecule expressed at sites of activation in vivo and inducible in cultured human umbilical vein endothelial cells (HUVEC) by treatment with cytokines such as recombinant human interleukin 1β (IL-1β). Liposomes of various types (classical, sterically stabilized, cationic, pH-sensitive), each conjugated with mAb H18/7, a murine monoclonal antibody that recognizes the extracellular domain of E-selectin, bound selectively and specifically to IL-1β-activated HUVEC at levels up to 275-fold higher than to unactivated HUVEC. E-selectin-targeted immunoliposomes appeared in acidic, perinuclear vesicles 2–4 hr after binding to the cell surface, consistent with internalization via the endosome/lysosome pathway. Activated HUVEC incubated with E-selectin-targeted immunoliposomes, loaded with the cytotoxic agent doxorubicin, exhibited significantly decreased cell survival, whereas unactivated HUVEC were unaffected by such treatment. These results demonstrate the feasibility of exploiting cell surface activation markers for the endothelial-selective delivery of biologically active agents via immunoliposomes. Application of this targeting approach in vivo may lead to novel therapeutic strategies in the treatment of cardiovascular disease.

Sustained secretion of human alpha-1-antitrypsin from murine muscle transduced with adeno-associated virus vectors

Recombinant adeno-associated virus (AAV) vectors have been used to transduce murine skeletal muscle as a platform for secretion of therapeutic proteins. The utility of this approach for treating alpha-1-antitrypsin (AAT) deficiency was tested in murine myocytes in vitro and in vivo. AAV vectors expressing the human AAT gene from either the cytomegalovirus (CMV) promoter (AAV-C-AT) or the human elongation factor 1-α promoter (AAV-E-AT) were examined. In vitro in C2C12 murine myoblasts, the expression levels in transient transfections were similar between the two vectors. One month after transduction, however, the human elongation factor 1 promoter mediated 10-fold higher stable human AAT expression than the CMV promoter. In vivo transduction was performed by injecting doses of up to 1.4 × 1013 particles into skeletal muscles of several mouse strains (C57BL/6, BALB/c, and SCID). In vivo, the CMV vector mediated higher levels of expression, with sustained serum levels over 800 μg/ml in SCID and over 400 μg/ml in C57BL/6 mice. These serum concentrations are 100,000-fold higher than those previously observed with AAV vectors in muscle and are at levels which would be therapeutic if achieved in humans. High level expression was delayed for several weeks but was sustained for over 15 wk. Immune responses were dependent upon the mouse strain and the vector dosage. These data suggest that recombinant AAV vector transduction of skeletal muscle could provide a means for replacing AAT or other essential serum proteins but that immune responses may be elicited under certain conditions.

Neurotransplantation of magnetically labeled oligodendrocyte progenitors: Magnetic resonance tracking of cell migration and myelination

Demyelination is a common pathological finding in human neurological diseases and frequently persists as a result of failure of endogenous repair. Transplanted oligodendrocytes and their precursor cells can (re)myelinate axons, raising the possibility of therapeutic intervention. The migratory capacity of transplanted cells is of key importance in determining the extent of (re)myelination and can, at present, be evaluated only by using invasive and irreversible procedures. We have exploited the transferrin receptor as an efficient intracellular delivery device for magnetic nanoparticles, and transplanted tagged oligodendrocyte progenitor cells into the spinal cord of myelin-deficient rats. Cell migration could be easily detected by using three-dimensional magnetic resonance microscopy, with a close correlation between the areas of contrast enhancement and the achieved extent of myelination. The present results demonstrate that magnetic resonance tracking of transplanted oligodendrocyte progenitors is feasible; this technique has the potential to be easily extended to other neurotransplantation studies involving different precursor cell types.

Development of an epithelium-specific expression cassette with human DNA regulatory elements for transgene expression in lung airways

The efficient expression of therapeutic genes in target cells or tissues is an important component of efficient and safe gene therapy. Utilizing regulatory elements from the human cytokeratin 18 (K18) gene, including 5′ genomic sequences and one of its introns, we have developed a novel expression cassette that can efficiently express reporter genes, as well as the human cystic fibrosis transmembrane conductance regulator (CFTR) gene, in cultured lung epithelial cells. CFTR transcripts expressed from the native K18 enhancer/promoter include two alternative splicing products, due to the activation of two cryptic splice sites in the CFTR coding region. Modification of the K18 intron and CFTR cDNA sequences eliminated the cryptic splice sites without changing the CFTR amino acid sequence, and led to enhanced CFTR mRNA and protein expression as well as biological function. Transgenic expression analysis in mice showed that the modified expression cassette can direct efficient and epithelium-specific expression of the Escherichia coli LacZ gene in the airways of fetal lungs, with no detectable expression in lung fibroblasts or endothelial cells. This is the first expression cassette which selectively directs lung transgene expression for CFTR gene therapy to airway epithelia.

Regulated expression of the diphtheria toxin A chain by a tumor-specific chimeric transcription factor results in selective toxicity for alveolar rhabdomyosarcoma cells

Alveolar rhabdomyosarcoma (ARMS) cells often harbor one of two unique chromosomal translocations, either t(2;13)(q35;q14) or t(1;13)(p36;q14). The chimeric proteins expressed from these rearrangements, PAX3-FKHR and PAX7-FKHR, respectively, are potent transcriptional activators. In an effort to exploit these unique cancer-specific molecules to achieve ARMS-specific expression of therapeutic genes, we have studied the expression of a minimal promoter linked to six copies of a PAX3 DNA binding site, prs-9. In transient transfections, expression of the prs-9-regulated reporter genes was ≈250-fold higher than expression of genes lacking the prs-9 sequences in cell lines derived from ARMS, but remained at or below baseline levels in other cells. High expression of these prs-9-regulated genes was also observed in a cancer cell line that lacks t(2;13) but was stably transfected with a plasmid expressing PAX3-FKHR. Transfection of a plasmid containing the diphtheria toxin A chain gene regulated by prs-9 sequences (pA3–6PED) was selectively cytotoxic for PAX3-FKHR-expressing cells. This was shown by inhibition of gene expression from cotransfected plasmids and by direct cytotoxicity after transfected cells were isolated by cell sorting. Gene transfer of pA3–6PED may thus be useful as a cancer-specific treatment strategy for t(2;13)- or t(1;13)-positive ARMS. Furthermore, gene transfer of fusion protein-regulated toxin genes might also be applied to the treatment of other cancers that harbor cancer-specific chromosomal translocations involving transcription factors.

Hypoxic Regulation of Vascular Endothelial Growth Factor mRNA Stability Requires the Cooperation of Multiple RNA Elements

Vascular endothelial growth factor (VEGF) is a key regulator of developmental, physiological, and tumor angiogenesis. Upregulation of VEGF expression by hypoxia appears to be a critical step in the neovascularization of solid cancers. The VEGF mRNA is intrinsically labile, but in response to hypoxia the mRNA is stabilized. We have systematically analyzed the regions in the VEGF mRNA that are responsible for its lability under normoxic conditions and for stabilization in response to hypoxia. We find that the VEGF mRNA not only contains destabilizing elements in its 3′ untranslated region (3′UTR), but also contains destabilizing elements in the 5′UTR and coding region. Each region can independently promote mRNA degradation, and together they act additively to effect rapid degradation under normoxic conditions. Stabilization of the mRNA in response to hypoxia is completely dependent on the cooperation of elements in each of the 5′UTR, coding region, and 3′UTR. Combinations of any of two of these three regions were completely ineffective in responding to hypoxia, whereas combining all three regions allowed recapitulation of the hypoxic stabilization seen with the endogenous VEGF mRNA. We conclude that multiple regions in the VEGF mRNA cooperate both to ensure the rapid degradation of the mRNA under normoxic conditions and to allow stabilization of the mRNA in response to hypoxia. Our findings highlight the complexity of VEGF gene expression and also reveal a mechanism of gene regulation that could become the target for strategies of therapeutic intervention.

Selection of peptides that functionally replace a zinc finger in the Sp1 transcription factor by using a yeast combinatorial library

We have developed a strategy for the identification of peptides able to functionally replace a zinc finger domain in a transcription factor. This strategy could have important ramifications for basic research on gene regulation and for the development of therapeutic agents. In this study in yeast, we expressed chimeric proteins that included a random peptide combinatorial library in association with two zinc finger domains and a transactivating domain. The library was screened for chimeric proteins capable of activating transcription from a target sequence in the upstream regulatory regions of selectable or reporter genes. In a screen of approximately 1.5 × 107 transformants we identified 30 chimeric proteins that exhibited transcriptional activation, some of which were able to discriminate between wild-type and mutant DNA targets. Chimeric library proteins expressed as glutathione S-transferase fusions bound to double-stranded oligonucleotides containing the target sequence, suggesting that the chimeras bind directly to DNA. Surprisingly, none of the peptides identified resembled a zinc finger or other well-known transcription factor DNA binding domain.

Statistical analysis of shape of objects based on landmark data.

Two objects with homologous landmarks are said to be of the same shape if the configurations of landmarks of one object can be exactly matched with that of the other by translation, rotation/reflection, and scaling. The observations on an object are coordinates of its landmarks with reference to a set of orthogonal coordinate axes in an appropriate dimensional space. The origin, choice of units, and orientation of the coordinate axes with respect to an object may be different from object to object. In such a case, how do we quantify the shape of an object, find the mean and variation of shape in a population of objects, compare the mean shapes in two or more different populations, and discriminate between objects belonging to two or more different shape distributions. We develop some methods that are invariant to translation, rotation, and scaling of the observations on each object and thereby provide generalizations of multivariate methods for shape analysis.

Inhibition of neoplastic development in the liver by hepatocyte growth factor in a transgenic mouse model.

Overexpression of the c-myc oncogene is associated with a variety of both human and experimental tumors, and cooperation of other oncogenes and growth factors with the myc family are critical in the evolution of the malignant phenotype. The interaction of hepatocyte growth factor (HGF) with c-myc during hepatocarcinogenesis in a transgenic mouse model has been analyzed. While sustained overexpression of c-myc in the liver leads to cancer, coexpression of HGF and c-myc in the liver delayed the appearance of preneoplastic lesions and prevented malignant conversion. Furthermore, tumor promotion by phenobarbital was completely inhibited in the c-myc/HGF double transgenic mice, whereas phenobarbital was an effective tumor promoter in the c-myc single transgenic mice. The results indicate that HGF may function as a tumor suppressor during early stages of liver carcinogenesis, and suggest the possibility of therapeutic application for this cytokine.

Efficient expression of protein coding genes from the murine U1 small nuclear RNA promoters.

Few promoters are active at high levels in all cells. Of these, the majority encode structural RNAs transcribed by RNA polymerases I or III and are not accessible for the expression of proteins. An exception are the small nuclear RNAs (snRNAs) transcribed by RNA polymerase II. Although snRNA biosynthesis is unique and thought not to be compatible with synthesis of functional mRNA, we have tested these promoters for their ability to express functional mRNAs. We have used the murine U1a and U1b snRNA gene promoters to express the Escherichia coli lacZ gene and the human alpha-globin gene from either episomal or integrated templates by transfection, or infection into a variety of mammalian cell types. Equivalent expression of beta-galactosidase was obtained from < 250 nucleotides of 5'-flanking sequence containing the complete promoter of either U1 snRNA gene or from the 750-nt cytomegalovirus promoter and enhancer regions. The mRNA was accurately initiated at the U1 start site, efficiently spliced and polyadenylylated, and localized to polyribosomes. Recombinant adenovirus containing the U1b-lacZ chimeric gene transduced and expressed beta-galactosidase efficiently in human 293 cells and airway epithelial cells in culture. Viral vectors containing U1 snRNA promoters may be an attractive alternative to vectors containing viral promoters for persistent high-level expression of therapeutic genes or proteins.

Self-renewal of primitive human hematopoietic cells (long-term-culture-initiating cells) in vitro and their expansion in defined medium.

A major goal of experimental and clinical hematology is the identification of mechanisms and conditions that support the expansion of transplantable hematopoietic stem cells. In normal marrow, such cells appear to be identical to (or represent a subset of) a population referred to as long-term-culture-initiating cells (LTC-ICs) so-named because of their ability to produce colony-forming cell (CFC) progeny for > or = 5 weeks when cocultured with stromal fibroblasts. Some expansion of LTC-ICs in vitro has recently been described, but identification of the factors required and whether LTC-IC self-renewal divisions are involved have remained unresolved issues. To address these issues, we examined the maintenance and/or generation of LTC-ICs from single CD34+ CD38- cells cultured for variable periods under different culture conditions. Analysis of the progeny obtained from cultures containing a feeder layer of murine fibroblasts engineered to produce steel factor, interleukin (IL)-3, and granulocyte colony-stimulating factor showed that approximately 20% of the input LTC-ICs (representing approximately 2% of the original CD34+ CD38- cells) executed self-renewal divisions within a 6-week period. Incubation of the same CD34+ CD38- starting populations as single cells in a defined (serum free) liquid medium supplemented with Flt-3 ligand, steel factor, IL-3, IL-6, granulocyte colony-stimulating factor, and nerve growth factor resulted in the proliferation of initial cells to produce clones of from 4 to 1000 cells within 10 days, approximately 40% of which included > or = 1 LTC-IC. In contrast, in similar cultures containing methylcellulose, input LTC-ICs appeared to persist but not divide. Overall the LTC-IC expansion in the liquid cultures was 30-fold in the first 10 days and 50-fold by the end of another 1-3 weeks. Documentation of human LTC-IC self-renewal in vitro and identification of defined conditions that permit their extensive and rapid amplification should facilitate analysis of the molecular mechanisms underlying these processes and their exploitation for a variety of therapeutic applications.

Suppression of experimental arthritis by gene transfer of interleukin 1 receptor antagonist cDNA.

Restoration of the impaired balance between pro- and antiinflammatory cytokines should provide effective treatment of rheumatoid arthritis. Gene therapy has been proposed as an approach for delivery of therapeutic proteins to arthritic joints. Here, we examined the efficacy of antiinflammatory gene therapy in bacterial cell wall-induced arthritis in rats. Human secreted interleukin 1 receptor antagonist (sIL-1ra) was expressed in joints of rats with recurrent bacterial cell wall-induced arthritis by using ex vivo gene transfer. To achieve this, primary synoviocytes were transduced in culture with a retroviral vector carrying the sIL-1ra cDNA. Transduced cells were engrafted in ankle joints of animals prior to reactivation of arthritis. Animals in control groups were engrafted with synoviocytes transduced with lacZ and neo marker genes. Cells continued to express transferred genes for at least 9 days after engraftment. We found that gene transfer of sIL-1ra significantly suppressed the severity of recurrence of arthritis, as assessed by measuring joint swelling and by the gross-observation score, and attenuated but did not abolish erosion of cartilage and bone. The effect of intraarticularly expressed sIL-1ra was essentially local, as there was no significant difference in severity of recurrence between unengrafted contralateral joints in control and experimental groups. We estimate that locally expressed sIL-1ra was about four orders of magnitude more therapeutically efficient than systemically administered recombinant sIL-1ra protein. These findings provide experimental evidence for the feasibility of antiinflammatory gene therapy for arthritis.