Dynamic slate recommendation with gated recurrent units and Thompson sampling

We consider the problem of recommending relevant content to users of an internet platform in the form of lists of items, called slates. We introduce a variational Bayesian Recurrent Neural Net recommender system that acts on time series of interactions between the internet platform and the user, and which scales to real world industrial situations. The recommender system is tested both online on real users, and on an offline dataset collected from a Norwegian web-based marketplace, FINN.no, that is made public for research. This is one of the first publicly available datasets which includes all the slates that are presented to users as well as which items (if any) in the slates were clicked on. Such a data set allows us to move beyond the common assumption that implicitly assumes that users are considering all possible items at each interaction. Instead we build our likelihood using the items that are actually in the slate, and evaluate the strengths and weaknesses of both approaches theoretically and in experiments. We also introduce a hierarchical prior for the item parameters based on group memberships. Both item parameters and user preferences are learned probabilistically. Furthermore, we combine our model with bandit strategies to ensure learning, and introduce ‘in-slate Thompson sampling’ which makes use of the slates to maximise explorative opportunities. We show experimentally that explorative recommender strategies perform on par or above their greedy counterparts. Even without making use of exploration to learn more effectively, click rates increase simply because of improved diversity in the recommended slates.

A tutorial on bridge sampling

The marginal likelihood plays an important role in many areas of Bayesian statistics such as parameter estimation, model comparison, and model averaging. In most applications, however, the marginal likelihood is not analytically tractable and must be approximated using numerical methods. Here we provide a tutorial on bridge sampling (Bennett, 1976; Meng & Wong, 1996), a reliable and relatively straightforward sampling method that allows researchers to obtain the marginal likelihood for models of varying complexity. First, we introduce bridge sampling and three related sampling methods using the beta-binomial model as a running example. We then apply bridge sampling to estimate the marginal likelihood for the Expectancy Valence (EV) model-a popular model for reinforcement learning. Our results indicate that bridge sampling provides accurate estimates for both a single participant and a hierarchical version of the EV model. We conclude that bridge sampling is an attractive method for mathematical psychologists who typically aim to approximate the marginal likelihood for a limited set of possibly high-dimensional models.

Overcoming indecision by changing the decision boundary

The dominant theoretical framework for decision making asserts that people make decisions by integrating noisy evidence to a threshold. It has recently been shown that in many ecologically realistic situations, decreasing the decision boundary maximizes the reward available from decisions. However, empirical support for decreasing boundaries in humans is scant. To investigate this problem, we used an ideal observer model to identify the conditions under which participants should change their decision boundaries with time to maximize reward rate. We conducted 6 expanded-judgment experiments that precisely matched the assumptions of this theoretical model. In this paradigm, participants could sample noisy, binary evidence presented sequentially. Blocks of trials were fixed in duration, and each trial was an independent reward opportunity. Participants therefore had to trade off speed (getting as many rewards as possible) against accuracy (sampling more evidence). Having access to the actual evidence samples experienced by participants enabled us to infer the slope of the decision boundary. We found that participants indeed modulated the slope of the decision boundary in the direction predicted by the ideal observer model, although we also observed systematic deviations from optimality. Participants using suboptimal boundaries do so in a robust manner, so that any error in their boundary setting is relatively inexpensive. The use of a normative model provides insight into what variable(s) human decision makers are trying to optimize. Furthermore, this normative model allowed us to choose diagnostic experiments and in doing so we present clear evidence for time-varying boundaries. (PsycINFO Database Record

Drosophila PINK1 and parkin loss-of-function mutants display a range of non-motor Parkinson's disease phenotypes

Parkinson's disease (PD) is more commonly associated with its motor symptoms and the related degeneration of dopamine (DA) neurons. However, it is becoming increasingly clear that PD patients also display a wide range of non-motor symptoms, including memory deficits and disruptions of their sleep-wake cycles. These have a large impact on their quality of life, and often precede the onset of motor symptoms, but their etiology is poorly understood. The fruit fly Drosophila has already been successfully used to model PD, and has been used extensively to study relevant non-motor behaviours in other contexts, but little attention has yet been paid to modelling non-motor symptoms of PD in this genetically tractable organism. We examined memory performance and circadian rhythms in flies with loss-of-function mutations in two PD genes: PINK1 and parkin. We found learning and memory abnormalities in both mutant genotypes, as well as a weakening of circadian rhythms that is underpinned by electrophysiological changes in clock neurons. Our study paves the way for further work that may help us understand the mechanisms underlying these neglected aspects of PD, thus identifying new targets for treatments to address these non-motor problems specifically and perhaps even to halt disease progression in its prodromal phase.

Context-dependent decision-making: a simple Bayesian model

Many phenomena in animal learning can be explained by a context-learning process whereby an animal learns about different patterns of relationship between environmental variables. Differentiating between such environmental regimes or 'contexts' allows an animal to rapidly adapt its behaviour when context changes occur. The current work views animals as making sequential inferences about current context identity in a world assumed to be relatively stable but also capable of rapid switches to previously observed or entirely new contexts. We describe a novel decision-making model in which contexts are assumed to follow a Chinese restaurant process with inertia and full Bayesian inference is approximated by a sequential-sampling scheme in which only a single hypothesis about current context is maintained. Actions are selected via Thompson sampling, allowing uncertainty in parameters to drive exploration in a straightforward manner. The model is tested on simple two-alternative choice problems with switching reinforcement schedules and the results compared with rat behavioural data from a number of T-maze studies. The model successfully replicates a number of important behavioural effects: spontaneous recovery, the effect of partial reinforcement on extinction and reversal, the overtraining reversal effect, and serial reversal-learning effects.

Posterior weighted reinforcement learning with state uncertainty

Reinforcement learning models generally assume that a stimulus is presented that allows a learner to unambiguously identify the state of nature, and the reward received is drawn from a distribution that depends on that state. However, in any natural environment, the stimulus is noisy. When there is state uncertainty, it is no longer immediately obvious how to perform reinforcement learning, since the observed reward cannot be unambiguously allocated to a state of the environment. This letter addresses the problem of incorporating state uncertainty in reinforcement learning models. We show that simply ignoring the uncertainty and allocating the reward to the most likely state of the environment results in incorrect value estimates. Furthermore, using only the information that is available before observing the reward also results in incorrect estimates. We therefore introduce a new technique, posterior weighted reinforcement learning, in which the estimates of state probabilities are updated according to the observed rewards (e.g., if a learner observes a reward usually associated with a particular state, this state becomes more likely). We show analytically that this modified algorithm can converge to correct reward estimates and confirm this with numerical experiments. The algorithm is shown to be a variant of the expectation-maximization algorithm, allowing rigorous convergence analyses to be carried out. A possible neural implementation of the algorithm in the cortico-basal-ganglia-thalamic network is presented, and experimental predictions of our model are discussed.

Challenges of functional imaging research of pain in children

Functional imaging has revolutionized the neurosciences. In the pain field it has dramatically altered our understanding of how the brain undergoes significant functional, anatomical and chemical changes in patients with chronic pain. However, most studies have been performed in adults. Because functional imaging is non-invasive and can be performed in awake individuals, applications in children have become more prevalent, but only recently in the pain field. Measures of changes in the brains of children have important implications in understanding neural plasticity in response to acute and chronic pain in the developing brain. Such findings may have implications for treatments in children affected by chronic pain and provide novel insights into chronic pain syndromes in adults. In this review we summarize this potential and discuss specific concerns related to the imaging of pain in children.

Serum lipids regulate dendritic cell CD1 expression and function

Dendritic cells (DCs) are highly potent antigen-presenting cells (APCs) and play a vital role in stimulating naïve T cells. Treatment of human blood monocytes with the cytokines granulocyte-macrophage colony-stimulating factor (GM-CSF) and interleukin (IL)-4 stimulates them to develop into immature dendritic cells (iDCs) in vitro. DCs generated by this pathway have a high capacity to prime and activate resting T cells and prominently express CD1 antigen-presenting molecules on the cell surface. The presence of human serum during the differentiation of iDCs from monocytes inhibits the expression of CD1a, CD1b and CD1c, but not CD1d. Correspondingly, T cells that are restricted by CD1c showed poor responses to DCs that were generated in the presence of human serum, while the responses of CD1d-restricted T cells were enhanced. We chemically fractionated human serum to isolate the bioactive factors that modulate surface expression of CD1 proteins during monocyte to DC differentiation. The human serum components that affected CD1 expression partitioned with polar organic soluble fractions. Lysophosphatidic acid and cardiolipin were identified as lipids present in normal human serum that potently modulate CD1 expression. Control of CD1 expression was mediated at the level of gene transcription and correlated with activation of the peroxisome proliferator-activated receptor (PPAR) nuclear hormone receptors. These findings indicate that the ability of human DCs to present lipid antigens to T cells through expression of CD1 molecules is sensitively regulated by lysophosphatidic acid and cardiolipin in serum, which are ligands that can activate PPAR transcription factors.

CD1-mediated gamma/delta T cell maturation of dendritic cells

Immature myeloid dendritic cells (DCs) express only low levels of major histocompatibility complex (MHC) class II but express high levels of CD1 a, b, and c antigen-presenting molecules at the cell surface. As Vdelta1+ gamma/delta T cells are the main tissue subset of gamma/delta T cells and they are known to recognize CD1c in the absence of specific foreign antigen recognition, we examined the possible interaction of these T cells with immature DCs. We show that CD1-restricted gamma/delta T cells can mediate the maturation of DCs. DC maturation required cell-cell contact and could be blocked by antibodies against CD1c. The maturation process was partially mediated by tumor necrosis factor alpha. Importantly, immature DCs matured in the presence of lipopolysaccharide and CD1-restricted gamma/delta T cells produced bioactive interleukin-12p70. In addition, these DCs were able to efficiently present peptide antigens to naive CD4+ T cells. CD1-restricted gamma/delta T cell recognition of immature DCs provides the human immune system with the capacity to rapidly generate a pool of mature DCs early during microbial invasion. This may be an important source of critical host signals for T helper type 1 polarization of antigen-specific naive T cells and the subsequent adaptive immune response.

CD1-dependent dendritic cell instruction

Both microbial products and T cell factors influence dendritic cell (DC) maturation. However, it is not known which T cells are capable of interacting with DCs at the initiation of adaptive immunity, when foreign antigen-specific T cells are rare. We show here that self-reactive CD1-restricted T cells can promote DC maturation by recognizing CD1 in the absence of foreign antigens. T cell recognition of all four CD1 isoforms can trigger DC maturation, but their distinct mechanisms of costimulation lead to profound differences in concomitant interleukin 12 p70 production. Distinct CD1-reactive T cells may thus differentially direct DC development early in the immune response, thereby controlling subsequent polarization of acquired immunity.

Evidence for human CD4+ T cells in the CD1-restricted repertoire: derivation of mycobacteria-reactive T cells from leprosy lesions

Both the CD4-CD8- (double negative) and CD4-CD8+ T cell lineages have been shown to contain T cells which recognize microbial lipid and glycolipid Ags in the context of human CD1 molecules. To determine whether T cells expressing the CD4 coreceptor could recognize Ag in the context of CD1, we derived CD4+ T cell lines from the lesions of leprosy patients. We identified three CD4+ Mycobacterium leprae-reactive, CD1-restricted T cell lines: two CD1b restricted and one CD1c restricted. These T cell lines recognize mycobacterial Ags, one of which has not been previously described for CD1-restricted T cells. The response of CD4+ CD1-restricted T cells, unlike MHC class II-restricted T cells, was not inhibited by anti-CD4 mAb, suggesting that the CD4 coreceptor does not impact positive or negative selection of CD1-restricted T cells. The CD4+ CD1-restricted T cell lines produced IFN-gamma and GM-CSF, the Th1 pattern of cytokines required for cell-mediated immunity against intracellular pathogens, but no detectable IL-4. The existence of CD4+ CD1-restricted T cells that produce a Th1 cytokine pattern suggests a contributory role in immunity to mycobacterial infection.

Self-recognition of CD1 by gamma/delta T cells: implications for innate immunity

The specificity of immunoglobulins and alpha/beta T cell receptors (TCRs) provides a framework for the molecular basis of antigen recognition. Yet, evolution has preserved a separate lineage of gamma/delta antigen receptors that share characteristics of both immunoglobulins and alpha/beta TCRs but whose antigens remain poorly understood. We now show that T cells of the major tissue gamma/delta T cell subset recognize nonpolymorphic CD1c molecules. These T cells proliferated in response to CD1+ presenter cells, lysed CD1c+ targets, and released T helper type 1 (Th1) cytokines. The CD1c-reactive gamma/delta T cells were cytotoxic and used both perforin- and Fas-mediated cytotoxicity. Moreover, they produced granulysin, an important antimicrobial protein. Recognition of CD1c was TCR mediated, as recognition was transferred by transfection of the gamma/delta TCR. Importantly, all CD1c-reactive gamma/delta T cells express V delta 1 TCRs, the TCR expressed by most tissue gamma/delta T cells. Recognition by this tissue pool of gamma/delta T cells provides the human immune system with the capacity to respond rapidly to nonpolymorphic molecules on professional antigen presenting cells (APCs) in the absence of foreign antigens that may activate or eliminate the APCs. The presence of bactericidal granulysin suggests these cells may directly mediate host defense even before foreign antigen-specific T cells have differentiated.

A combination of two immunotoxins exerts synergistic cytotoxic activity against human breast-cancer cell lines

In previous studies, combinations of immunotoxins reactive with different cell-surface antigens have exerted additive cytotoxicity against tumor cells in culture. In this report we describe a combination of 2 immunotoxins that produce synergistic cytotoxic activity. Recombinantly derived ricin A chain (RTA) was conjugated with murine monoclonal antibodies (MAbs) 317G5, 260F9, 454A12 and 741F8 that bound to cell-surface determinants of 42, 55, 180 (transferrin receptor) and 185 kDa (HER-2/neu) expressed by the SKBr3 human breast-cancer cell line. When inhibition of clonogenic growth was measured in a limiting dilution assay, the combination of 260F9-RTA and 454A12-RTA produced synergistic cytotoxic activity against SKBr3 and 2 other breast-cancer cell lines. All other combinations produced only additive inhibition of clonogenic growth. Simultaneous binding of 260F9 and 454A12 was not supra-additive, but sub-populations of cells which lacked one or the other antigen could be detected. Kinetic studies of internalization, using antibodies conjugated with gold particles, indicated that 454A12 remained within peripheral endosomes for a longer interval in the presence of 260F9. This change in the traffic of the transferrin receptor may contribute to synergy between 260F9-RTA and 454A12-RTA.

Detection of breast carcinoma cells in human bone marrow using fluorescence-activated cell sorting and conventional cytology

Flow cytometry has been compared with conventional cytology as a method for detecting breast carcinoma cells in human bone marrow. Breast cancer cells from patient effusion fluids or from established cell lines were mixed with normal human bone marrow at dilutions ranging from 1:10 to 1:100,000. Cells were labeled with five directly fluoresceinated murine monoclonal antibodies reactive against epithelial cell surface determinants of 42, 55, 72, 200, and more than 200 kD. Fluorescence of tumor cells within the mixtures was then analyzed with the use of a fluorescence-activated cell sorter. In multiple experiments, one tumor cell in 10,000 nucleated marrow cells could be detected by flow cytometry. In addition, a linear correlation was observed over approximately 3 logs between the number of tumor cells added and the number of tumor cells detected. In a double-blind study comparing flow cytometry and cytology, flow cytometric analysis detected one tumor cell among 10,000 marrow precursors in 14 of 15 instances, whereas standard cytologic methods detected similar tumor contamination in 9 of 15 instances. Neither technique used individually could detect one tumor cell in 100,000 bone marrow cells. Used in combination, however, flow cytometry and cytology could detect one breast cancer cell among 100,000 normal marrow progenitors.

Use of immunotoxins in combination to inhibit clonogenic growth of human breast carcinoma cells

Substantial heterogeneity has been observed in the expression of individual antigens within tumor cell populations. Immunotoxins which bind to different cell surface antigens might exert additive or synergistic cytotoxicity when used in combination to eliminate all clonogenic cells within a tumor. Immunotoxins have been prepared by conjugating recombinantly derived toxin A chain to different monoclonal reagents which recognize cell surface determinants of Mr 42,000 (317G5), 55,000 (260F9), and 200,000 (741F8). Each immunotoxin was evaluated for binding, internalization, and cytotoxicity with four breast cancer cell lines. Each of the three immunotoxins bound to the SKBr3 cell line and exerted antitumor activity in a limiting dilution clonogenic assay. Simultaneous treatment with two immunotoxins produced additive antitumor activity with each of the possible combinations. Additive binding could be demonstrated by immunofluorescent techniques, however, with only one of three combinations. With two of the three combinations, subpopulations of tumor cells could be identified which lacked one or the other antigenic determinant but not both. Consequently, log-additive antitumor activity was produced by immunotoxins in combination, and heterogeneity of antigenic targets may have contributed to the combined cytotoxicity in some but not all cases.

Heterogeneity of antigen expression in advanced epithelial ovarian cancer

Immunohistochemical techniques were used to evaluate the expression of six antigens (CA 125, TAG 72, CA 19-9, OVTL3, DF3, and transferrin receptor) in frozen sections from the primary tumor and metastases of 20 patients with epithelial ovarian cancer. Heterogeneous expression of most antigens was observed within a given tumor nodule, but in each patient the proportion of cells expressing an antigen was similar in the primary tumor and metastases. To explore the stability of the antigenic phenotype of individual cells, we studied CA 125 expression in an ovarian cancer cell line. Cells were separated into CA 125-positive and -negative groups using fluorescence-activated cell sorting. After the two groups of cells were recultured separately, only 38% of cells originally sorted as CA 125 positive still expressed CA 125, whereas 27% of cells sorted as CA 125 negative expressed CA 125. That cells may gain or lose CA 125 expression in culture suggests that expression of CA 125 by ovarian cancer cells is not a stable trait.

Elimination of malignant clonogenic breast cancer cells from human bone marrow

Autologous bone marrow transplantation is a promising approach to the treatment of breast cancer but is at present limited to patients without bone marrow metastases. To eliminate malignant clonogenic breast cancer cells from normal human bone marrow, immunomagnetic separation has been combined with chemoseparation using 4-hydroperoxycyclophosphamide. Breast cancer cell lines have been mixed with a 10-fold excess of irradiated human bone marrow from normal donors. Mixtures have been incubated with a combination of five different monoclonal antibodies which bind to epithelial cell surface antigens of Mr 42,000, 55,000, 72,000, 200,000, and greater than 200,000. Antiglobulin coated microspheres which contained magnetite were added, and tumor cells were trapped in a magnetic field. Elimination of tumor cells from the decanted marrow was measured in a limiting dilution assay. Two treatments with antibody and microspheres permitted elimination of 2-4 logs of clonogenic breast cancer cells, depending upon the cell line studied. Similar treatment of nonirradiated normal marrow failed to affect levels of colony forming units-granulocyte-macrophage significantly. Use of immunomagnetic purging in combination with 4-hydroperoxycyclophosphamide eliminated up to 5 logs of tumor cells but reduced the recovery of colony forming units-granulocyte-macrophage. If prompt engraftment is observed following reinfusion of similarly treated marrow in phase I trials, these techniques should permit extension of autologous bone marrow transplantation to a larger population of breast cancer patients.

Differential induction by interferons of major histocompatibility complex-encoded and non-major histocompatibility complex-encoded antigens in human breast and ovarian carcinoma cell lines

Treatment of cancer cells with interferons can modulate expression of cell surface antigens, particularly those of the major histocompatibility complex (MHC). To examine the effect of recombinant gamma- and alpha-interferons on expression of non-MHC antigens, murine monoclonal antibodies have been used to quantitate 14 distinct tumor-associated cell surface antigens from five breast cancer cell lines and five ovarian cancer cell lines using a live cell radioimmunoassay. Both Class I and Class II MHC antigens could be augmented or induced with gamma-interferon. Significantly increased expression of MHC antigens was observed in nine of 10 cell lines with induction indices as high as 11-fold. When 17 non-MHC epitopes were measured on 10 cell lines, minimal (1.3-2.7-fold) induction was observed in 10 of the 170 instances evaluated. Expression of only two epitopes, 2G3 and 735B11, was increased on more than one cell line. On six cell lines expression of non-MHC epitopes could not be increased. Consequently, among many different cell surface determinants, interferons produced a highly selective augmentation or induction of MHC antigens, whereas augmentation or induction of other tumor-associated antigens was apparently restricted to a few epitopes.

Human postnatal CD4- CD8- CD3- thymic T cell precursors differentiate in vitro into T cell receptor delta-bearing cells

The signals required for activation and the differentiation of human triple negative postnatal thymocytes were studied in vitro. Highly purified populations of CD4-, CD8-, CD3- (triple negative) thymocytes were isolated by combined panning and preparative cell sorting and the ability of triple negative thymocytes to proliferate in response to various cytokines determined. Maximal triple negative proliferation was obtained using a mitogenic combination of CD2 antibodies and either rIL-2 or the phorbol ester, PMA. Long term growth (2 to 6 wk) of postnatal triple negative thymocytes was best achieved using CD2 antibodies and rIL-2. After in vitro culture with CD2 antibodies and rIL-2, triple negative thymocytes gave rise to TCR-delta+ cells beginning on day 2 of culture (approximately 15% CD3/TCR-delta+) reaching maximum (approximately 60% CD3/TCR-delta+) on day 7 with stable number of TCR-delta+ cells observed in vitro for up to 6 wk. Analysis of 30 clones of human postnatal triple negative thymocytes demonstrated 9 of 30 (30%) were TCR-delta+, beta F1-, essentially ruling out overgrowth of the triple negative population over time by a minor pool of contaminating TCR-delta+ cells. Thus, these studies have defined an in vitro culture system for human postnatal T cell precursors and demonstrated that precursors of human TCR-gamma delta+ T cells reside in the triple negative thymocyte pool.

Human postnatal CD4- CD8- CD3- thymic T cell precursors differentiate in vitro into T cell receptor delta-bearing cells

The signals required for activation and the differentiation of human triple negative postnatal thymocytes were studied in vitro. Highly purified populations of CD4-, CD8-, CD3- (triple negative) thymocytes were isolated by combined panning and preparative cell sorting and the ability of triple negative thymocytes to proliferate in response to various cytokines determined. Maximal triple negative proliferation was obtained using a mitogenic combination of CD2 antibodies and either rIL-2 or the phorbol ester, PMA. Long term growth (2 to 6 wk) of postnatal triple negative thymocytes was best achieved using CD2 antibodies and rIL-2. After in vitro culture with CD2 antibodies and rIL-2, triple negative thymocytes gave rise to TCR-delta+ cells beginning on day 2 of culture (approximately 15% CD3/TCR-delta+) reaching maximum (approximately 60% CD3/TCR-delta+) on day 7 with stable number of TCR-delta+ cells observed in vitro for up to 6 wk. Analysis of 30 clones of human postnatal triple negative thymocytes demonstrated 9 of 30 (30%) were TCR-delta+, beta F1-, essentially ruling out overgrowth of the triple negative population over time by a minor pool of contaminating TCR-delta+ cells. Thus, these studies have defined an in vitro culture system for human postnatal T cell precursors and demonstrated that precursors of human TCR-gamma delta+ T cells reside in the triple negative thymocyte pool.

Combined chemoseparation and immunoseparation of clonogenic T lymphoma cells from human bone marrow using 2'-deoxycoformycin, deoxyadenosine, 3A1 monoclonal antibody, and complement

Chemoseparation and immunoseparation techniques have been combined to eliminate malignant clonogenic T lymphoma cells from human bone marrow. Incubation with 5 microM 2'-deoxycoformycin and 500 microM deoxyadenosine has eliminated 2 logs of HSB-2 T lymphoma cells from a 20-fold excess of irradiated human bone marrow. Multiple incubations with 3A1 antibody and rabbit complement eliminated approximately 2 logs of HSB-2 cells from similar mixtures. Used in combination, the 2 techniques eliminated up to 4 logs of T lymphoma cells. Incubation of normal human bone marrow under similar conditions failed to affect growth of granulocyte-macrophage colony-forming cell units, burst-forming erythroid units, or multipotential erythroid-granulocyte-megakaryocyte-macrophage colony-forming hematopoietic progenitor cells units.

Hydroxyurea synchronization increases mitotic yield in human glioma cell lines

Karyotypic studies of human gliomas are often limited by a low mitotic index and the appearance of contracted chromosomes that do not exhibit clear banding patterns. The purpose of this study was to investigate the use of hydroxyurea (HU) as a synchronizing agent using established human glioma cell lines as a model system. HU was shown to reversibly inhibit cellular replication in glioma cell lines U-251MG, D-245MG, D-247MG and D-263MG by flow cytometry on the basis of DNA content. Two-to sixfold increases were demonstrated in the mitotic index of HU-treated cultures exhibiting the greatest percentage of cells in the G2/M phases of the cell cycle. HU, therefore, promises to be an effective agent for use in short term cultures from biopsied human tumors to increase the quality of chromosome preparations in these tumors.