This paper 1 aims to quantitatively explain rationales of each prediction that is made by a pre-trained convolutional neural network (CNN). We propose to learn a decision tree, which clarifies the specific reason for each prediction made by the CNN at the semantic level. I.e. the decision tree decomposes feature representations in high conv-layers of the CNN into elementary concepts of object parts. In this way, the decision tree tells people which object parts activate which filters for the prediction and how much they contribute to the prediction score. Such semantic and quantitative explanations for CNN predictions have specific values beyond the traditional pixel-level analysis of CNNs. More specifically, our method mines all potential decision modes of the CNN, where each mode represents a common case of how the CNN uses object parts for prediction. The decision tree organizes all potential decision modes in a coarse-tofine manner to explain CNN predictions at different finegrained levels. Experiments have demonstrated the effectiveness of the proposed method.

In recent years, a broad range of nanocrystals have been synthesized in droplet-based microfluidic reactors which provide obvious advantages, such as accurate manipulation, better reproducibility and reliable automation. In this review, we initially introduce general concepts of droplet reactors followed by discussions of their main functional regions including droplet generation, mixing of reactants, reaction controlling, in situ monitoring, and reaction quenching. Subsequently, the enhanced mass and heat transport properties are discussed. Next, we focus on research frontiers including sequential multistep synthesis, intelligent synthesis, reliable scale-up synthesis, and interfacial synthesis. Finally, we end with an outlook on droplet reactors, especially highlighting some aspects such as large-scale production, the integrated process of synthesis and post-synthetic treatments, automated droplet reactors with in situ monitoring and optimizing algorithms, and rapidly developing strategies for interfacial synthesis.

Bone regeneration is very important for the recovery of some diseases including osteoporosis and bone fracture trauma. It is a multiple-step- and multiple-gene-involved complex process, including the matrix secretion and calcium mineralization by osteoblasts differentiated from mesenchymal stem cells (MSCs) and the absorption of calcium and phosphorus by osteoclasts differentiated from hematopoietic stem cells. Long noncoding RNAs (lncRNAs) are a family of transcripts longer than 200 nt without or with very low protein-coding potential. Recent studies have demonstrated that lncRNAs are widely involved in the regulation of lineage commitment and differentiation of stem cells through multiple mechanisms. In this review, we will summarize the roles and molecular mechanism of lncRNAs including H19, MALAT1, MODR, HOTAIR, DANCR, MEG3, HoxA-AS3, and MIAT in osteogenesis ossification; lncRNA ZBED3-AS1 and CTA-941F9.9, DANCR, and HIT in chondrogenic differentiation; and lncRNA DANCR in osteoclast differentiation. These findings will facilitate the development and application of novel molecular drugs which regulate the balance of bone formation and absorption.

Long noncoding RNAs (lncRNAs) have been demonstrated to be important regulators during the osteogenic differentiation of mesenchymal stem cells (MSCs). We analyzed the lncRNA expression profile during osteogenic differentiation of human umbilical cord-derived mesenchymal stem cells (hUC-MSCs) and identified a significantly downregulated lncRNA RP11-527N22.2, named osteogenic differentiation inhibitory lncRNA 1, ODIR1. In hUC-MSCs, ODIR1 knockdown significantly promoted osteogenic differentiation, whereas overexpression inhibited osteogenic differentiation in vitro and in vivo. Mechanistically, ODIR1 interacts with F-box protein 25 (FBXO25) and facilitates the proteasome-dependent degradation of FBXO25 by recruiting Cullin 3 (CUL3). FBXO25 increases the mono-ubiquitination of H2BK120 (H2BK120ub) which subsequently promotes the trimethylation of H3K4 (H3K4me3). Both H2BK120ub and H3K4me3 form a loose chromatin structure, inducing the transcription of the key transcription factor osterix (OSX) and increasing the expression of the downstream osteoblast markers, osteocalcin (OCN), osteopontin (OPN), and alkaline phosphatase (ALP). In summary, ODIR1 acts as a key negative regulator during the osteogenic differentiation of hUC-MSCs through the FBXO25/H2BK120ub/H3K4me3/OSX axis, which may provide a novel understanding of lncRNAs that regulate the osteogenesis of MSCs and a potential therapeutic strategy for the regeneration of bone defects.

Supramolecular chirality and its complete self‐recovery ability are highly mystical in nature and biological systems, which remains a major challenge today. Herein, we demonstrate that partially cross‐linked azobenzene (Azo) units can be employed as the potential chiral trigger to fully heal the destroyed helical superstructure in achiral nematic polymer system. Combining the self‐assembly of Azo units and terminal hydroxyl groups in polymer side chains allows the vapor‐induced chiral nematic phase and covalent fixation of the superstructure via acetal reaction. The induced helical structure of Azo units can be stored by inter‐chain cross‐linking, even after removal of the chiral source. Most interestingly, the stored chiral information can trigger perfect chiral self‐recovery (CSR) behavior after being destroyed by UV light, heat, and solvents. The results pave a new way for producing novel chiroptical materials with reversible chirality from achiral sources.

No abstract

No abstract

Strain modulation is crucial for heteroepitaxy such as GaN on foreign substrates. Here, the epitaxy of strain-relaxed GaN films on graphene/SiC substrates by metal-organic chemical vapor deposition is demonstrated. Graphene was directly prepared on SiC substrates by thermal decomposition. Its pre-treatment with nitrogen-plasma can introduce C–N dangling bonds, which provides nucleation sites for subsequent epitaxial growth. The scanning transmission electron microscopy measurements confirm that part of graphene surface was etched by nitrogen-plasma. We study the growth behavior on different areas of graphene surface after pre-treatment, and propose a growth model to explain the epitaxial growth mechanism of GaN films on graphene. Significantly, graphene is found to be effective to reduce the biaxial stress in GaN films and the strain relaxation improves indium-atom incorporation in InGaN/GaN multiple quantum wells (MQWs) active region, which results in the obvious red-shift of light-emitting wavelength of InGaN/GaN MQWs. This work opens up a new way for the fabrication of GaN-based long wavelength light-emitting diodes.