Independently Interpretable Lasso for Generalized Linear Models.

Sparse regularization such as ℓ1 regularization is a quite powerful and widely used strategy for high-dimensional learning problems. The effectiveness of sparse regularization has been supported practically and theoretically by several studies. However, one of the biggest issues in sparse regularization is that its performance is quite sensitive to correlations between features. Ordinary ℓ1 regularization selects variables correlated with each other under weak regularizations, which results in deterioration of not only its estimation error but also interpretability. In this letter, we propose a new regularization method, independently interpretable lasso (IILasso), for generalized linear models. Our proposed regularizer suppresses selecting correlated variables, so that each active variable affects the response independently in the model. Hence, we can interpret regression coefficients intuitively, and the performance is also improved by avoiding overfitting. We analyze the theoretical property of the IILasso and show that the proposed method is advantageous for its sign recovery and achieves almost minimax optimal convergence rate. Synthetic and real data analyses also indicate the effectiveness of the IILasso.

Culture-independent analysis of fecal microbiota in infants, with special reference to Bifidobacterium species.

Fecal microbiota of 31 breast-fed, 26 mix-fed, and 11 bottle-fed infants were analyzed by using terminal restriction fragment length polymorphism (T-RFLP), and culture method. We first determined the total and cultivated bacterial counts in infant fecal microbiota. Only approximately 30% of bacteria present in fecal microbiota were cultivable while the remainder was yet-to-be cultured bacteria. Sixty-eight fecal samples were divided into two clusters (I and II) by T-RFLP analysis, and then subdivided into five subclusters (Ia, Ib, IIa, IIb and IIc). There was no clear relationship between clusters and feeding method. A proportion of bifidobacteria was detected in the fecal material by PCR method using species-specific primers. The predominant Bifidobacterium spp. was Bifidobacterium longum longum type (43 samples (63.2%)), followed by B. longum infantis type (23 samples (33.8%)) and B. breve (16 samples (23.5%)). The distribution of Bifidobacterium spp. was similar in the three feeding groups. In contrast, the high incidence of B. breve in cluster I, especially subcluster Ia and B. longum longum type in cluster II, especially subcluster IIa and IIc were characterized by T-RFLP method. Our results showed that the colonization of Bifidobacterium spp. in infant feces correlated with the T-RFLP clusters.

[Retinopathy of prematurity in extremely low birth weight infants: a Tokyo multicenter study].

OBJECTIVE: To investigate how the increase in survival rate in extremely low birth weight (a birth weight of 1,000 g or less) infants had affected the incidence of retinopathy of prematurity (ROP) and the frequency of laser treatment. METHODS: We retrospectively reviewed the medical records of 122 surviving premature infants with birthweights less than 1,000 g to determine the severity of ROP observed at 16 neonatal intensive care units in Tokyo between April and October 2002. RESULTS: The survival rate was 85.6%. The mean gestational age was 26.74 weeks and the mean birth weight was 782.25 g. One-hundred-and-five infants (86.1%) developed ROP, fifty (41.0%) received laser treatment, and six (4.9%) had retinal detachment. The median postmenstrual age (gestational age at birth plus chronological age in weeks, PMA) at the onset of ROP was 32.5 weeks, and the first laser treatment was performed at the median PMA of 35.7 weeks. CONCLUSIONS: In these extremely low birth weight infants, there was an increase in the survival rate and in the incidence of severe ROP that progressed to the stage that required treatment.

Detection and quantification of four species of the genus Clostridium in infant feces.

To determine the composition of Clostridium in the feces of infants approximately 30 days old, we have developed a detection and quantification method of Clostridium paraputrificum, Clostridium perfringens, Clostridium tertium, and Clostridium difficile by species-specific primers. C. perfringens and C. difficile were detected in four fecal samples from 22 infants (18.2%), whereas C. paraputrificum was detected in three samples (16.7%). C. tertium was detected in two samples (9.1%). Moreover, the occurrences of the four species in bottle-and mix-fed infants were relatively higher than in breast-fed infants (P< 0.05). Subsequently, positive samples detected by nested PCR (polymerase chain reaction) were subjected to realtime PCR. The results showed that the numbers of C. paraputrificum, C. perfringens, C. tertium, and C. difficile ranged from about 1x10(5) to 3x10(7) cells/g wet feces.