Jérôme Paul

Kernel methods for heterogeneous feature selection

This paper introduces two feature selection methods to deal with heterogeneous data that include continuous and categorical variables. We propose to plug a dedicated kernel that handles both kinds of variables into a Recursive Feature Elimination procedure using either a non-linear SVM or Multiple Kernel Learning. These methods are shown to offer state-of-the-art performances on a variety of high-dimensional classification tasks.

Clinical Parameters vs Cytokine Profiles as Predictive Markers of IgE-Mediated Allergy in Young Children.

Background Allergy afflicts one third of children, negatively impacting their quality of life and generating a significant socio-economic burden. To this day, this disorder remains difficult to diagnose early in young patients, with no predictive test available. Objective This study was designed to correlate cytokine profiles with clinical phenotypes of allergy development. Methods Three hundred patients were recruited and followed from birth to 18 months of age. They were given a clinical exam at birth and at 2, 6, 12, and 18 months of age, with skin prick tests at 6, and 18 months, in order to have a record of their medical history and determine their allergic status. In addition, mononuclear cells from 131 patients were isolated from cord blood and from peripheral blood samples at 2, 6 and 18 months of age, to analyse their cytokine and chemokine production. Results Cord blood mononuclear cells (CBMCs) from future Immunoglobulin (Ig) E-mediated allergic children produced significantly less Interleukin (IL)-12p70 and IL-15 than cells from the rest of the cohort. Multivariate analyses revealed that the best predictive model of allergy was built on cytokine data, whereas the best predictive model of IgE-mediated allergy was built on clinical parameters. Conclusions and clinical relevance Although univariate analyses can yield interesting information regarding the immune responses of allergic children, finding predictive markers of the disorder will likely rely on monitoring multiple parameters. Nonetheless these analyses suggest a potential key role for IL-15 in the development of atopic disease. In addition, the study highlights the importance of clinical parameters in predicting the development of IgE-mediated allergy.

Long-term Outcomes with Anti-TNF Therapy and Accelerated Step-up in the Prospective Pediatric Belgian Crohn's Disease Registry (BELCRO)

Background: Accelerated step-up or anti–tumor necrosis factor (TNF) before first remission is currently not recommended in pediatric Crohns disease. Methods: Five-year follow-up data from a prospective observational cohort of children diagnosed with Crohns disease in Belgium were analyzed. Disease severity was scored as inactive, mild, or moderate to severe. Remission or inactive disease was defined as sustained if lasting ≥2 years. Univariate analyses were performed between anti-TNF–exposed versus naive patients and anti-TNF before versus after first remission and correlations assessed with primary outcomes average disease severity and sustained remission. Results: A total of 91 patients (median [IQR] age 12.7 [10.9–14.8] yrs, 53% male) were included. Disease location was 12% L1, 23% L2, and 64% L3 with 76% upper gastrointestinal and 30% perianal involvement. Disease severity was 25% mild and 75% moderate to severe. Of 66 (73%) anti-TNF–exposed patients, 34 (52%) had accelerated step-up. Anti-TNF use was associated with age (13.1 [11.5–15.2] versus 11.8 [8.7–13.8] yrs; P < 0.05), L2 (29% versus 8%; P = 0.04), and average disease severity (1.7 [1.4–1.9] versus 1.4 [1.3–1.6]; P < 0.001). Duration of anti-TNF correlated with average disease severity (r = 0.32, P = 0.002). Accelerated step-up was also associated with age (13.3 [12.1–15.9] versus 12.5 [10.2–14.1]; P = 0.02) and average disease severity (1.8 [1.6–1.9] versus 1.6 [1.3–1.8]; P = 0.002). Duration of sustained remission was similar in all patients, and no serious infections, cancer, or deaths were reported. Conclusions: Anti-TNF therapy and accelerated step-up in older patients with more severe disease leads to beneficial long-term outcomes.

Inferring statistically significant features from random forests

Embedded feature selection can be performed by analyzing the variables used in a Random Forest. Such a multivariate selection takes into account the interactions between variables but is not straightforward to interpret in a statistical sense. We propose a statistical procedure to measure variable importance that tests if variables are significantly useful in combination with others in a forest. We show experimentally that this new importance index correctly identifies relevant variables. The top of the variable ranking is largely correlated with Breiman׳s importance index based on a permutation test. Our measure has the additional benefit to produce p-values from the forest voting process. Such p-values offer a very natural way to decide which features are significantly relevant while controlling the false discovery rate. Practical experiments are conducted on synthetic and real data including low and high-dimensional datasets for binary or multi-class problems. Results show that the proposed technique is effective and outperforms recent alternatives by reducing the computational complexity of the selection process by an order of magnitude while keeping similar performances.