How do you test a program when only a single user, with no expertise in software testing, is able to determine if the program is performing correctly? Such programs are common today in the form of machine-learned classifiers. We consider the problem of testing this common kind of machine-generated program when the only oracle is an end user: e.g., only you can determine if your email is properly filed. We present test selection methods that provide very good failure rates even for small test suites, and show that these methods work in both large-scale random experiments using a “gold standard” and in studies with real users. Our methods are inexpensive and largely algorithm-independent. Key to our methods is an exploitation of properties of classifiers that is not possible in traditional software testing. Our results suggest that it is plausible for time-pressured end users to interactively detect failures-even very hard-to-find failures-without wading through a large number of successful (and thus less useful) tests. We additionally show that some methods are able to find the arguably most difficult-to-detect faults of classifiers: cases where machine learning algorithms have high confidence in an incorrect result.

Over the past two decades, tissue engineering and regenerative medicine have evolved from what many considered a theoretical science to what is now a clinical reality. Tissue engineering combines biomaterial scaffolds, growth factors and stem or progenitor cells to repair damaged tissues. Adipose tissue, an abundant and easily accessed tissue, is a potential source of stromal/stem cells for regenerative therapeutic applications. Like bone marrow-derived mesenchymal stem cells, adipose-derived stromal/stem cells display both immunomodulatory and immunosuppressive properties. The adipose cells exert these actions, in part, through their secretion of paracrine growth factors. This review highlights recent developments in the isolation, characterization and preclinical application of adipose-derived cells and the challenges facing their translation into clinical practice.

When intelligent interfaces, such as intelligent desktop assistants, email classifiers, and recommender systems, customize themselves to a particular end user, such customizations can decrease productivity and increase frustration due to inaccurate predictions—especially in early stages when training data is limited. The end user can improve the learning algorithm by tediously labeling a substantial amount of additional training data, but this takes time and is too ad hoc to target a particular area of inaccuracy. To solve this problem, we propose new supervised and semi-supervised learning algorithms based on locally-weighted logistic regression for feature labeling by end users, enabling them to point out which features are important for a class, rather than provide new training instances.

Adipose-derived mesenchymal stem or stromal cells (ASCs) are poised for clinical use in an allogeneic setting. Although ASCs have been shown to be nonimmunogenic by several laboratories, it is advisable for the investigator to confirm this for ASCs used in their studies due to variations in ASC production and the animal models in which they are used. We describe here the use of the mixed lymphocyte reaction (MLR) assay to determine immunogenicity and suppression by ASCs in vitro as well as assessing T cell responses to allogeneic ASC transplantation in vivo. A flow cytometry assay to determine serum antibody titer to transplanted ASCs is also described.

Applications that adapt to a particular end user often make inaccurate predictions during the early stages when training data is limited. Although an end user can improve the learning algorithm by labeling more training data, this process is time consuming and too ad hoc to target a particular area of inaccuracy. To solve this problem, we propose a new learning algorithm based on Locally Weighted Logistic Regression for feature labeling by end users, enabling them to point out which features are important for a class, rather than provide new training instances. In our user study, the first allowing ordinary end users to freely choose features to label directly from text documents, our algorithm was more effective than others at leveraging end users’ feature labels to improve the learning algorithm. Our results strongly suggest that allowing users to freely choose features to label is a promising method for allowing end users to improve learning algorithms effectively.

When intelligent interfaces, such as intelligent desktop assistants, email classifiers, and recommender systems, customize themselves to a particular end user, such customizations can decrease productivity and increase frustration due to inaccurate predictions - especially in early stages, when training data is limited. The end user can improve the learning algorithm by tediously labeling a substantial amount of additional training data, but this takes time and is too ad hoc to target a particular area of inaccuracy. To solve this problem, we propose a new learning algorithm based on locally weighted regression for feature labeling by end users, enabling them to point out which features are important for a class, rather than provide new training instances. In our user study, the first allowing ordinary end users to freely choose features to label directly from text documents, our algorithm was both more effective than others at leveraging end users' feature labels to improve the learning algorithm, and more robust to real users' noisy feature labels. These results strongly suggest that allowing users to freely choose features to label is a promising method for allowing end users to improve learning algorithms effectively.

Intelligent assistants are handling increasingly critical tasks, but until now, end users have had no way to systematically assess where their assistants make mistakes. For some intelligent assistants, this is a serious problem: if the assistant is doing work that is important, such as assisting with qualitative research or monitoring an elderly parent's safety, the user may pay a high cost for unnoticed mistakes. This paper addresses the problem with WYSIWYT/ML (What You See Is What You Test for Machine Learning), a human/computer partnership that enables end users to systematically test intelligent assistants. Our empirical evaluation shows that WYSIWYT/ML helped end users find assistants' mistakes significantly more effectively than ad hoc testing. Not only did it allow users to assess an assistant's work on an average of 117 predictions in only 10 minutes, it also scaled to a much larger data set, assessing an assistant's work on 623 out of 1,448 predictions using only the users' original 10 minutes' testing effort.

Posterolateral spinal fusion is the standard treatment for lumbar compression fractures. Adult adipose tissue-derived stem cells (ASCs) promote osteogenesis in vivo and in vitro. The hypothesis tested in this study was that syngeneic and allogeneic ASCs on a biomaterial scaffold composed of tricalcium phosphate and collagen I will accelerate spinal fusion in a rat model. ASCs from male Fischer or ACI rats were loaded onto scaffolds (53,571 cells/mm(3)) and cultured in stromal media for 48 h. Male Fisher rats were assigned to 4 cohorts (n = 14/cohort) after bilateral decortication of the L4 and L5 transverse processes: (1) No treatment; (2) scaffold only; (3) scaffold + syngeneic ASCs; or (4) scaffold + allogeneic ASCs. Half of each cohort was harvested 4 or 8 weeks after surgery. Spinal fusion was evaluated with radiographs, microcomputed tomography, and light microscopy. Callus did not form in spines without scaffolds. There were no significant differences in callus formation among scaffold cohorts 4 weeks after surgery. Callus formation was more mature in both ASC cohorts versus scaffold alone 8 weeks after surgery based on microstructure as well as radiographic and microcomputed tomographic evidence of active bone formation. Inflammatory cell infiltrate was significantly lower in both ASC cohorts (syngeneic = 18.3 +/- 0.85%; allogeneic = 23.5 +/- 2.33%) versus scaffold alone (46.8 +/- 11.8%) 4 weeks after surgery. Results of this study support syngeneic and allogeneic ASC acceleration of posterior lumbar spinal fusion in a rat model.

Allogeneic transplantation of stem cells provides a cost-effective approach for their clinical utilization. Because of multiple methodologies used for the production and utilization of stem cells, a method to assess the immunogenicity of stem cells prior to and after preclinical studies in animals would be extremely beneficial. The use of the mixed lymphocyte reaction (MLR) assay is described for this purpose. The MLR assay is a technique in which responder cells containing T lymphocytes are mixed with allogeneic stimulator cells, and T-cell proliferation is measured as an index of reactivity. Variations of this assay are described in which stem cells are used as stimulators to determine immunogenicity, stem cells are added to MLR cultures to determine suppression, and T cells obtained from animals treated with stem cells are cultured with cells from the stem cell donor to determine whether the animals were primed to the stem cell allograft.

Many applications include machine learning algorithms intended to learn “programs” (rules of behavior) from an end user’s actions. When these learned programs are wrong, their users receive little explanation as to why, and even less freedom of expression to help the machine learn from its mistakes. In this paper, we develop and explore a set of candidate principles for providing salient debugging information to end users who would like to correct these programs. We informed the candidate principles through a formative study, built a prototype that instantiates them, and conducted a user study of the prototype to collect empirical evidence to inform future variants. Our results suggest the value of exposing the machine’s reasoning process, supporting a flexible debugging vocabulary, and illustrating the effects of user changes to the learned program’s logic. Author

Adipose-derived stem cells (ASCs) express a nonimmunogenic profile as shown by in vitro studies that demonstrate a lack of T cell proliferation to allogeneic ASCs as well as ASC-mediated suppression of mixed lymphocyte reactions. To determine whether these observations would translate in vivo, immune monitoring studies were carried out in conjunction with a rat spinal fusion study. ASCs derived from Fischer or ACI strain rats were loaded onto scaffolds and implanted in Fischer recipients that had undergone the following treatments: (1) No treatment; (2) Scaffold only; (3) Syngeneic ASCs + Scaffold; or (4) Allogeneic ASCs + Scaffold. Half of each group was sacrificed at 4 weeks postimplantation, and the remaining animals were sacrificed at 8 weeks. As determined in a separate study, allogeneic and syngeneic ASCs were equally efficacious in accelerating spinal fusion compared to No treatment and Scaffold only control groups. To determine whether donor ASCs induced an immune response in recipient rats, lymph nodes were harvested for T cell proliferation studies and serum was collected to assess antibody responses. Although T cell priming was not detected to donor alloantigens in recipients at either time point, significant antibody responses were detected to ACI ASCs in animals implanted with syngeneic or allogeneic ASCs. Antibodies were of the IgG isotype, noncytotoxic in the presence of complement, and reactive to fetal bovine serum. These results support the use of allogeneic ASCs for spinal fusion.

Bone marrow-derived stromal cells (BMSCs) exhibit extraordinary degree of plasticity and growth factor repertoire for which they have been investigated for repair and regeneration of damaged tissues, but have not been adequately examined for wound healing. The ability of BMSCs to accelerate healing of surgically inflicted cutaneous and fascial wounds was tested in vivo in rats and in vitro using a fibroblast monolayer wound model. Intravenous treatment with BMSCs augmented healing of both cutaneous and fascial wounds as determined by an increase in the biomechanical strength of wounds. In vitro experiments showed that incorporation of BMSCs in fibroblast monolayers accelerates the closure of mechanically disrupted monolayers, which was attributed to the enhanced migration of fibroblasts onto the denuded surfaces. Furthermore, culture medium conditioned by activated BMSCs promoted the closure of defects in monolayers and enhanced the proliferation/growth and directional migration (chemotaxis) of fibroblasts. This study demonstrates that BMSCs significantly augment healing of cutaneous and fascial wounds in vivo at least in part through interaction with fibroblasts in which BMSCs promote growth and chemotaxis of fibroblasts.

Cells isolated from Wharton's jelly, referred to as umbilical cord matrix stromal (UCMS) cells, adhere to a tissue-culture plastic substrate, express mesenchymal stromal cell (MSC) surface markers, self-renew, and are multipotent (differentiate into bone, fat, cartilage, etc.) in vitro. These properties support the notion that UCMS cells are a member of the MSC family. Here, the immune properties of UCMS cells are characterized in vitro. The overall hypothesis is that UCMS cells possess immune properties that would be permissive to allogeneic transplantation. For example, UCMS cells will suppress of the proliferation of "stimulated" lymphocytes (immune suppression) and have reduced immunogenicity (e.g., would be poor stimulators of allogeneic lymphocyte proliferation). Hypothesis testing was as follows: first, the effect on proliferation of coculture of mitotically inactivated human UCMS cells with concanavalin-A-stimulated rat splenocytes was assessed in three different assays. Second, the effect of human UCMS cells on one-way and two-way mixed lymphocyte reaction (MLR) assays was determined. Third, the expression of human leukocyte antigen (HLA)-G was examined in human UCMS cells using reverse transcription-polymerase chain reaction, since HLA-G expression conveys immune regulatory properties at the maternal-fetal interface. Fourth, the expression of CD40, CD80, and CD86 was determined by flow cytometry. Fifth, the cytokine expression of UCMS cells was evaluated by focused gene array. The results indicate that human UCMS cells inhibit splenocyte proliferation response to concanavalin A stimulation, that they do not stimulate T-cell proliferation in a one-way MLR, and that they inhibit the proliferation of stimulated T cells in a two-way MLR. Human UCMS cells do not inhibit nonstimulated splenocyte proliferation, suggesting specificity of the response. UCMS cells express mRNA for pan-HLA-G. UCMS cells do not express the costimulatory surface antigens CD40, CD80, and CD86. UCMS cells express vascular endothelial growth factor and interleukin-6, molecules previously implicated in the immune modulation observed in MSCs. In addition, the array data indicate that UCMS cells make a cytokine and other factors that may support hematopoiesis. Together, these results support previous observations made following xenotransplantation; for example, there was no evidence of frank immune rejection of undifferentiated UCMS cells. The results suggest that human UCMS will be tolerated in allogeneic transplantation. Disclosure of potential conflicts of interest is found at the end of this article.

Introduction: Back pain afflicts over 75% of Americans during their lifetime. Posterolateral spinal fusion is the standard treatment for lumbar compression fractures at an annual direct medical cost of over $746 million in the US. Failure rates associated with spinal fusion are as high as 44% and increase costs substantially. Higher success rates will have significant economic, medical and societal implications. Bone grafts are used to facilitate and accelerate spinal fusion. Autografts are the current “gold standard”, but there are complications associated with graft harvest. Regenerative medicine employs stem cells to promote tissue healing. Adipose tissue is an abundant and accessible source of adipose tissue-derived stromal cells (ASCs) which promote osteogenesis both in vivo and in vitro. This study was designed to test the hypothesis that syngeneic and allogeneic ASCs combined with a biomaterial scaffold will accelerate spinal fusion compared to no treatment or scaffold alone in a rat model. Materials and Methods: Subcutaneous adipose tissue was harvested from 10 week old male Fischer and ACI rats (n=16/cohort). Tissue was minced, washed, and suspended in phosphate buffered saline containing 1% bovine serum albumin and 0.1% collagenase type I. Following a 60-minute digestion at 37°C, the suspension was centrifuged. The pelleted stromal vascular fraction cells were plated in stromal media (DMEM/F-12 Ham’s Media with 10% fetal bovine serum and 1% antibiotic/antimycotic) and incubated in a humidified 5% CO2 incubator to 75% confluency. The ASCs were then harvested and expanded up to 2 passages. Cell aliquots (5 x 106 cells/ml) were cryopreserved in liquid nitrogen. Cell fractions were evaluated in vitro for their osteogenic and adipogenic capacity using standard assays. Prior to surgery, cells were defrosted, rinsed, suspended in stromal media and loaded onto 0.5 x 0.2 x 0.7 cm scaffold blocks (Vitoss, Orthovita, Malvern, PA) blocks. Blocks loaded with cells or media alone were cultured in stromal media for 48 hours prior to surgery. Sixty-four male Fisher rats (200g) were randomly assigned to one of 4 cohorts (n=16/cohort): 1) No treatment; 2) Scaffold only; 3) Scaffold + syngeneic ASCs; or 4) Scaffold + allogeneic ASCs. Following humane euthanasia, spines were harvested from half of each cohort 4 or 8 weeks after surgery. The surgical procedure was performed according to published methods. Rats were anesthetized with isoflurane. The lumbar region was aseptically prepared, and the posterior lumbar spine was exposed. The L4 and L5 transverse processes were decorticated bilaterally. Following decortication, routine closure was performed in the no treatment cohort while scaffold blocks with or without ASCs were placed at the level of the L4-L5 intervertebral space prior to closure in scaffold cohorts. Cell loading on the blocks was confirmed with scanning electron microscopy (SEM). Following harvest, spines were imaged with micro-CT. They were then fixed in formalin, decalcified and embedded in paraffin. Coronal sections at the level of the L4-L5 transverse processes were stained with H&E and Masson’s trichrome. The quality and quantity of spinal fusion was evaluated with radiographs, micro-computed tomography (micro-CT), and light microscopy. Outcome measures were statistically evaluated with MANOVA models that tested the four group comparisons multivariately. Significance was considered at p < .05. Results: All animals survived the surgical procedure and post-operative period. Cells were evident on loaded scaffolds prior to surgical implantation with SEM. The cells demonstrated both adipogenic and osteogenic potential in vitro. Spines without scaffolds did not have evidence of callus formation 4 or 8 weeks after surgery. There was radiographic and micro-CT evidence of callus formation at the level of the L4-L5 intervertebral space in all three scaffold treatment groups 4 and 8 weeks after surgery. Based on light and polarized light microscopy callus composition was primarily fibrocartilagenous, and there were no significant differences in outcome measures between scaffold cohorts 4 weeks after surgery. Callus formation was more highly organized in the ASC cohorts compared to the scaffold only cohort based on radiographic and micro-CT evaluation eight weeks after surgery. The observations were further confirmed by callus calcification evident with light and polarized light microscopy in ASC cohorts. There was no evidence of callus calcification in the scaffold only cohort 8 weeks after surgery.

Adipose tissue represents an abundant and accessible source of multipotent adult stem cells and is used by many investigators for tissue engineering applications; however, not all laboratories use cells at equivalent stages of isolation and passage. We have compared the immunophenotype of freshly isolated human adipose tissue-derived stromal vascular fraction (SVF) cells relative to serial-passaged adipose-derived stem cells (ASCs). The initial SVF cells contained colony-forming unit fibroblasts at a frequency of 1:32. Colony-forming unit adipocytes and osteoblasts were present in the SVF cells at comparable frequencies (1:28 and 1:16, respectively). The immunophenotype of the adipose-derived cells based on flow cytometry changed progressively with adherence and passage. Stromal cell-associated markers (CD13, CD29, CD44, CD63, CD73, CD90, CD166) were initially low on SVF cells and increased significantly with successive passages. The stem cell-associated marker CD34 was at peak levels in the SVF cells and/or early-passage ASCs and remained present, although at reduced levels, throughout the culture period. Aldehyde dehydrogenase and the multidrug-resistance transport protein (ABCG2), both of which have been used to identify and characterize hematopoietic stem cells, are expressed by SVF cells and ASCs at detectable levels. Endothelial cell-associated markers (CD31, CD144 or VE-cadherin, vascular endothelial growth factor receptor 2, von Willebrand factor) were expressed on SVF cells and did not change significantly with serial passage. Thus, the adherence to plastic and subsequent expansion of human adipose-derived cells in fetal bovine serum-supplemented medium selects for a relatively homogeneous cell population, enriching for cells expressing a stromal immunophenotype, compared with the heterogeneity of the crude SVF.