Instance Consistency Regularization for Semi-Supervised 3D Instance Segmentation.

Large-scale datasets with point-wise semantic and instance labels are crucial to 3D instance segmentation but also expensive. To leverage unlabeled data, previous semi-supervised 3D instance segmentation approaches have explored self-training frameworks, which rely on high-quality pseudo labels for consistency regularization. They intuitively utilize both instance and semantic pseudo labels in a joint learning manner. However, semantic pseudo labels contain numerous noise derived from the imbalanced category distribution and natural confusion of similar but distinct categories, which leads to severe collapses in self-training. Motivated by the observation that 3D instances are non-overlapping and spatially separable, we ask whether we can solely rely on instance consistency regularization for improved semi-supervised segmentation. To this end, we propose a novel self-training network InsTeacher3D to explore and exploit pure instance knowledge from unlabeled data. We first build a parallel base 3D instance segmentation model DKNet, which distinguishes each instance from the others via discriminative instance kernels without reliance on semantic segmentation. Based on DKNet, we further design a novel instance consistency regularization framework to generate and leverage high-quality instance pseudo labels. Experimental results on multiple large-scale datasets show that the InsTeacher3D significantly outperforms prior state-of-the-art semi-supervised approaches.

High-performance grating-like SERS substrate based on machine learning for ultrasensitive detection of Zexie-Baizhu decoction.

Rapid, universal and accurate identification of chemical composition changes in multi-component traditional Chinese medicine (TCM) decoction is a necessary condition for elucidating the effectiveness and mechanism of pharmacodynamic substances in TCM. In this paper, SERS technology, combined with grating-like SERS substrate and machine learning method, was used to establish an efficient and sensitive method for the detection of TCM decoction. Firstly, the grating-like substrate prepared by magnetron sputtering technology was served as a reliable SERS sensor for the identification of TCM decoction. The enhancement factor (EF) of 4-ATP probe molecules was as high as 1.90 × 107 and the limit of detection (LOD) was as low as 1 × 10-10 M. Then, SERS technology combined with support vector machine (SVM), decision tree (DT), Naive Bayes (NB) and other machine learning algorithms were used to classify and identify the three TCM decoctions, and the classification accuracy rate was as high as 97.78 %. In summary, it is expected that the proposed method combining SERS and machine learning method will have a high development in the practical application of multi-component analytes in TCM.

Environmental sustainability assessment of a new food waste anaerobic digestion and pyrolysis hybridization system.

This research conducted an environmental life cycle assessment (LCA) to evaluate an anaerobic digestion-co-pyrolysis (ADCo-Py) system in which pyrolysis was added to traditional food waste (FW) anaerobic digestion (AD) systems to treat the solid fraction and impurities separated from FW. The solid fraction, including impurities such as wooden chopsticks, plastics, eggshells, and bones, is usually incinerated, while pyrolysis can be a viable alternative to optimize FW treatment. The environmental impact of ADCo-Py was compared with stand-alone AD, pyrolysis, and ADCo-INC (AD with incineration of separated solids). The results indicated that both ADCo-Py (-1.726 kg CO2-Eq/kgFW) and ADCo-INC (-1.535 kg CO2-Eq/kgFW) outperform stand-alone AD (-0.855 kg CO2-Eq/kgFW) and pyrolysis (-0.181 kg CO2-Eq/kgFW) in mitigating global warming potential (GWP). Additionally, pretreatments were found to have the most significant influence on GWP, ecotoxicity potential (ETP), and acidification potential (AP). The two-step pretreatment in ADCo-Py, including the separation of solids and drying, significantly improved the environmental sustainability of the system when compared with standalone pyrolysis.

Heteroatom substitution enhances generation and reactivity of surface-activated peroxydisulfate complexes for catalytic fenton-like reactions.

Peroxydisulfate (PDS)-based Fenton-like reactions are promising advanced oxidation processes (AOPs) to degrade recalcitrant organic water pollutants. Current research predominantly focuses on augmenting the generation of reactive species (e.g., surface-activated PDS complexes (PDS*) to improve treatment efficiency, but overlooks the potential benefits of enhancing the reactivity of these species. Here, we enhanced PDS* generation and reactivity by incorporating Zn into CuO catalyst lattice, which resulted in 99% degradation of 4-chlorophenol within only 10 min. Zn increased PDS* generation by nearly doubling PDS adsorption while maintaining similar PDS to PDS* conversion efficiency, and induced higher PDS* reactivity than the common catalyst CuO, as indicated by a 4.1-fold larger slope between adsorbed PDS and open circuit potential of a catalytic electrode. Cu-O-Zn formation upshifts the d-band center of Cu sites and lowers the energy barrier for PDS adsorption and sulfate desorption, resulting in enhanced PDS* generation and reactivity. Overall, this study informs strategies to enhance PDS* reactivity and design highly active catalysts for efficient AOPs.

Rerouting phytosterol degradation pathway for directed androst-1,4-diene-3,17-dione microbial bioconversion.

Steroid-based drugs are now mainly produced by the microbial transformation of phytosterol, and a two-step bioprocess is adopted to reach high space-time yields, but byproducts are frequently observed during the bioprocessing. In this study, the catabolic switch between the C19- and C22-steroidal subpathways was investigated in resting cells of Mycobacterium neoaurum NRRL B-3805, and a dose-dependent transcriptional response toward the induction of phytosterol with increased concentrations was found in the putative node enzymes including ChoM2, KstD1, OpccR, Sal, and Hsd4A. Aldolase Sal presented a dominant role in the C22 steroidal side-chain cleavage, and the byproduct was eliminated after sequential deletion of opccR and sal. Meanwhile, the molar yield of androst-1,4-diene-3,17-dione (ADD) was increased from 59.4 to 71.3%. With the regard of insufficient activity of rate-limiting enzymes may also cause byproduct accumulation, a chromosomal integration platform for target gene overexpression was established supported by a strong promoter L2 combined with site-specific recombination in the engineered cell. Rate-limiting steps of ADD bioconversion were further characterized and overcome. Overexpression of the kstD1 gene further strengthened the bioconversion from AD to ADD. After subsequential optimization of the bioconversion system, the directed biotransformation route was developed and allowed up to 82.0% molar yield with a space-time yield of 4.22 g·L-1·day-1. The catabolic diversion elements and the genetic overexpression tools as confirmed and developed in present study offer new ideas of M. neoaurum cell factory development for directed biotransformation for C19- and C22-steroidal drug intermediates from phytosterol. KEY POINTS: • Resting cells exhibited a catabolic switch between the C19- and C22-steroidal subpathways. • The C22-steroidal byproduct was eliminated after sequential deletion of opccR and sal. • Rate-limiting steps were overcome by promoter engineering and chromosomal integration.