To accelerate research on adversarial examples and robustness of machine learning classifiers, Google Brain organized a NIPS 2017 competition that encouraged researchers to develop new methods to generate adversarial examples as well as to develop new ways to defend against them. In this chapter, we describe the

We propose GraphNVP, the first invertible, normalizing flow-based molecular graph generation model. We decompose the generation of a graph into two steps: generation of (i) an adjacency tensor and (ii) node attributes. This decomposition yields the exact likelihood maximization on graph-structured data, combined with two novel reversible flows. We empirically demonstrate that our model efficiently generates valid molecular graphs with almost no duplicated molecules. In addition, we observe that the learned latent space can be used to generate molecules with desired chemical properties.

To clarify the interactive effect of the simultaneous death of dwarf bamboo (Sasa kurilensis), forest canopy gap formation, and seed predators on beech (Fagus crenata) regeneration, we analyzed beech demography from seed fall until the end of the first growing season of seedlings in an old-growth forest near Lake Towada, northern Japan. The simultaneous death of S. kurilensis took place in 1995. We established four types of sampling site differing in forest canopy conditions (closed or gap) and Sasa status (dead or alive). Beech seed survival and emergence ratio were both highest in gaps with dead Sasa (gap-dead), because rate of predation was lowest. Seedling survival during the first growing season was also highest in the gap-dead treatment, because of less predation and less damping off. As a result, even though density of seed fall was lowest in the gap-dead treatment, the living seedling density there was highest at the end of the first growing season. Predation, which caused the greatest mortality during the seed and seedling stages, was significantly lower at both sites in gaps and sites with dead Sasa. This was probably due to changes in the behavior of rodents in response to the structure of the forest canopy and undergrowth. Both the death of Sasa and canopy gap formation allowed seedlings to avoid damping off because of the high light availability. The indirect effect of the simultaneous death of Sasa and canopy gap formation in reducing predation contributed more to beech regeneration than their direct effect in increasing light for the seedlings.

SummaryMatrix vesicle-mediated mineralization is an orchestrated sequence of ultrastructural and biochemical events that lead to crystal nucleation and growth. The influx of phosphate ions into the matrix vesicle is mediated by several proteins such as TNAP, ENPP1, Pit1, annexin and so forth. The catalytic activity of ENPP1 generates pyrophosphate (PPi) using extracellular ATPs as a substrate, and the resultant PPi prevents crystal overgrowth. However, TNAP hydrolyzes PPi into phosphate ion monomers, which are then transported into the matrix vesicle through Pit1. Accumulation of Ca2+ and PO43− inside matrix vesicles then induces crystalline nucleation, with calcium phosphate crystals budding off radially, puncturing the matrix vesicle’s membrane and finally growing out of it to form mineralized nodules.

We hypothesized that the disordered tissue architecture in cancer results from the steps that the cells execute the program designed during ontogeny in a spatiotemporally inappropriate manner. HOX genes are known as master regulators of embryonic morphogenesis, and encode transcription factors which regulate the transcription of the downstream genes to realize the program of body plan. In this study, we quantified the expression levels of 39 HOX genes in 41 human non-small cell lung cancer (non-SCLC) and non-cancerous lung tissues by a comprehensive analysis system based on the realtime RT-PCR method. We found that the expression levels of HOXA1, A5, A10 and C6 in squamous cell carcinoma tissues (and HOXA5 and A10 in adenocarcinoma tissues) were significantly higher than those in the non-cancerous tissues. Comparison of HOX gene expressions between adenocarcinoma and squamous cell carcinoma tissues showed higher expressions of HOXA1, D9, D10 and D11 in squamous cell carcinoma tissues than in adenocarcinoma tissues. Immunohistochemical analysis revealed that HOXA5 and A10 proteins were localized in the cytoplasm of tumor cells in both adenocarcinoma and squamous cell carcinoma tissues. These results suggest that the disordered patterns of HOX gene expressions were involved in not only the development of non-SCLC but also histological diversity such as adenocarcinoma and 4 squamous cell carcinoma.