Genomic evidence for evolutionary history and local adaptation of two endemic apricots: Prunus hongpingensis and P. zhengheensis.

Apricot, belonging to the Armeniaca section of Rosaceae, is one of the economically important crop fruits that has been extensively cultivated. The natural wild apricots offer valuable genetic resources for crop improvement. However, some of them are endemic, with small populations, and are even at risk of extinction. In this study we unveil chromosome-level genome assemblies for two southern China endemic apricots, Prunus hongpingensis (PHP) and P. zhengheensis (PZH). We also characterize their evolutionary history and the genomic basis of their local adaptation using whole-genome resequencing data. Our findings reveal that PHP and PZH are closely related to Prunus armeniaca and form a distinct lineage. Both species experienced a decline in effective population size following the Last Glacial Maximum (LGM), which likely contributed to their current small population sizes. Despite the observed decrease in genetic diversity and heterozygosity, we do not observe an increased accumulation of deleterious mutations in these two endemic apricots. This is likely due to the combined effects of a low inbreeding coefficient and strong purifying selection. Furthermore, we identify a set of genes that have undergone positive selection and are associated with local environmental adaptation in PHP and PZH, respectively. These candidate genes can serve as valuable genetic resources for targeted breeding and improvement of cultivated apricots. Overall, our study not only enriches our comprehension of the evolutionary history of apricot species but also offers crucial insights for the conservation and future breeding of other endemic species amidst rapid climate changes.

Multilayer regulatory landscape and new regulators identification for bud dormancy release and bud break in Populus.

For trees originating from boreal and temperate regions, the dormancy-to-active transition, also known as bud dormancy release and bud break, are crucial processes that allow trees to reactive growth in the spring. The molecular mechanisms underlying these two processes remain poorly understood. Here, through integrative multiomics analysis of the transcriptome, DNA methylome, and proteome, we gained insights into the reprogrammed cellular processes associated with bud dormancy release and bud break. Our findings revealed multilayer regulatory landscapes governing bud dormancy release and bud break regulation, providing a valuable reference framework for future functional studies. Based on the multiomics analysis, we have determined a novel long intergenic noncoding RNA named Phenology Responsive Intergenic lncRNA 1 (PRIR1) plays a role in the activation of bud break. that the molecular mechanism of PRIR1 has been preliminary explored, and it may partially promote bud break by activating its neighbouring gene, EXORDIUM LIKE 5 (PtEXL5), which has also been genetically confirmed as an activator for bud break. This study has revealed a lncRNA-mediated regulatory mechanism for the control of bud break in Populus, operating independently of known regulatory pathways.

A WOX homolog disrupted by a transposon led to the loss of spines and contributed to the domestication of lettuce.

The loss of spines is one of the most important domestication traits for lettuce (Lactuca sativa). However, the genetics and regulation of spine development in lettuce remain unclear. We examined the genetics of spines in lettuce using a segregating population derived from a cross between cultivated and wild lettuce (Lactuca serriola). A gene encoding WUSCHEL-related homeobox transcription factor, named as WOX-SPINE1 (WS1), was identified as the candidate gene controlling the spine development in lettuce, and its function on spines was verified. A CACTA transposon was found to be inserted into the first exon of the ws1 allele, knocking out its function and leading to the lack of spines in cultivated lettuce. All lettuce cultivars investigated have the nonfunctional ws1 gene, and a selection sweep was found at the WS1 locus, suggesting its important role in lettuce domestication. The expression levels of WS1 were associated with the density of spines among different accessions of wild lettuce. At least two independent loss-of-function mutations in the ws1 gene caused the loss of spines in wild lettuce. These findings provide new insights into the development of spines and facilitate the exploitation of wild genetic resources in future lettuce breeding programs.

Fe-CGS Effectively Inhibits the Dynamic Migration and Transformation of Cadmium and Arsenic in Soil.

The iron-modified coal gasification slag (Fe-CGS) material has excellent performance in purifying heavy-metal-contaminated water due to its good surface properties and adsorption capacities. However, it is unclear whether it can provide long-term simultaneous stabilization of Cd and As in composite-contaminated soils in extreme environments. This study investigated the long-term stabilization of Cd and As in acidic (JLG) and alkaline (QD) soils by simulating prolonged heavy rainfall with the addition of Fe-CGS. Multiple extraction methods were used to analyze the immobilization mechanisms of Cd and As in soil and their effects on bioavailability. The results indicate that the stabilization efficiency was related to the dosage of Fe-CGS. The concentrations of Cd and As in the JLG soil leachate were reduced by 77.6% (2.0 wt%) and 87.8% (1.0 wt%), respectively. Additionally, the availability of Cd and As decreased by 46.7% (2.0 wt%) and 53.0% (1.0 wt%), respectively. In the QD soil leachate, the concentration of Cd did not significantly change, while the concentration of As decreased by 92.3% (2.0 wt%). Furthermore, the availability of Cd and As decreased by 22.1% (2.0 wt%) and 40.2% (1.0 wt%), respectively. Continuous extraction revealed that Fe-CGS facilitated the conversion of unstable, acid-soluble Cd into oxidizable Cd and acid-soluble Cd. Additionally, it promoted the transformation of both non-specifically and specifically adsorbed As into amorphous iron oxide-bound and residual As. Fe-CGS effectively improved the soil pH, reduced the bioavailability of Cd and As, and blocked the migration of Cd and As under extreme rainfall leaching conditions. It also promoted the transformation of Cd and As into more stable forms, exhibiting satisfactory long-term stabilization performance for Cd and As.

Unveiling unique alternative splicing responses to low temperature in Zoysia japonica through ZjRTD1.0, a high-quality reference transcript dataset.

Inadequate reference databases in RNA-seq analysis can hinder data utilization and interpretation. In this study, we have successfully constructed a high-quality reference transcript dataset, ZjRTD1.0, for Zoysia japonica, a widely-used turfgrass with exceptional tolerance to various abiotic stress, including low temperatures and salinity. This dataset comprises 113,089 transcripts from 57,143 genes. BUSCO analysis demonstrates exceptional completeness (92.4%) in ZjRTD1.0, with reduced proportions of fragmented (3.3%) and missing (4.3%) orthologs compared to prior datasets. ZjRTD1.0 enables more precise analyses, including transcript quantification and alternative splicing assessments using public datasets, which identified a substantial number of differentially expressed transcripts (DETs) and differential alternative splicing (DAS) events, leading to several novel findings on Z. japonica's responses to abiotic stresses. First, spliceosome gene expression influenced alternative splicing significantly under abiotic stress, with a greater impact observed during low-temperature stress. Then, a significant positive correlation was found between the number of differentially expressed genes (DEGs) encoding protein kinases and the frequency of DAS events, suggesting the role of protein phosphorylation in regulating alternative splicing. Additionally, our results suggest possible involvement of serine/arginine-rich (SR) proteins and heterogeneous nuclear ribonucleoproteins (hnRNPs) in generating inclusion/exclusion isoforms under low-temperature stress. Furthermore, our investigation revealed a significantly enhanced overlap between DEGs and differentially alternatively spliced genes (DASGs) in response to low-temperature stress, suggesting a unique co-regulatory mechanism governing transcription and splicing in the context of low-temperature response. In conclusion, we have proven that ZjRTD1.0 will serve as a reliable and useful resource for future transcriptomic analyses in Z. japonica.

Lactate drives CD38 signaling to promote Epithelial-Mesenchymal Transition through Snail induction in non-small cell lung cancer cells.

CD38 is the main NADase in mammalian cells. It regulates the homeostasis of nicotinamide adenine dinucleotide (NAD+) and extracellular nucleotides. Its function plays an important role in infection and aging. However, its potential functions in tumor cells have not been fully elucidated. In the present study, we demonstrated that lactate, which is derived from tumor metabolism remodeling, upregulates the expression of CD38 through OXPHOS-driven Hippo-TAZ pathway. The highly expressed CD38 converts NAD + to adenosine through the CD203a/CD73 complex and adenosine binds and activates its receptor A2AR, inducing the expression of Snail and promoting the invasion and metastasis of lung cancer cells. This finding elucidates a new perspective on the interplay between NAD + metabolism and glycolysis in tumor development.

Attributed Multi-Order Graph Convolutional Network for Heterogeneous Graphs.

Heterogeneous graph neural networks play a crucial role in discovering discriminative node embeddings and relations from multi-relational networks. One of the key challenges in heterogeneous graph learning lies in designing learnable meta-paths, which significantly impact the quality of learned embeddings. In this paper, we propose an Attributed Multi-Order Graph Convolutional Network (AMOGCN), which automatically explores meta-paths that involve multi-hop neighbors by aggregating multi-order adjacency matrices. The proposed model first constructs different orders of adjacency matrices from manually designed node connections. Next, AMOGCN fuses these various orders of adjacency matrices to create an intact multi-order adjacency matrix. This process is supervised by the node semantic information, which is extracted from the node homophily evaluated by attributes. Eventually, we employ a one-layer simplifying graph convolutional network with the learned multi-order adjacency matrix, which is equivalent to the cross-hop node information propagation with multi-layer graph neural networks. Substantial experiments reveal that AMOGCN achieves superior semi-supervised classification performance compared with state-of-the-art competitors.

Spatiotemporal Controllable Sono-Nanovaccines Driven by Free-Field Based Whole-Body Ultrasound for Personalized Cancer Therapy.

Therapeutic cancer vaccines fail to produce satisfactory outcomes against solid tumors since vaccine-induced anti-tumor immunity is significantly hampered by immunosuppression. Generating an in situ cancer vaccine targeting immunological cold tumor microenvironment (TME) appears attractive. Here, a type of free-field based whole-body ultrasound (US)-driven nanovaccines are constructed, named G5-CHC-R, by conjugating the sonosensitizer, Chenghai chlorin (CHC) and the immunomodulator, resiquimod (R848) on top of a super small-sized dendrimeric nanoscaffold. Once entering tumors, R848 can be cleaved from a hypoxia-sensitive linker, thus modifying the TME via converting macrophage phenotypes. The animals bearing orthotopic pancreatic cancer with intestinal metastasis and breast cancer with lung metastasis are treated with G5-CHC-R under a free-field based whole-body US system. Benefit from the deep penetration capacity and highly spatiotemporal selectiveness, G5-CHC-R triggered by US represented a superior alternative for noninvasive irradiation of deep-seated tumors and magnification of local immune responses via driving mass release of tumor antigens and "cold-warm-hot" three-state transformation of TME. In addition to irradiating primary tumors, a robust adaptive anti-tumor immunity is potentiated, leading to successful induction of systemic tumor suppression. The sono-nanovaccines with good biocompatibility posed wide applicability to a broad spectrum of tumors, revealing immeasurable potential for translational research in oncology.

New mechanistic insight into catalytic decomposition of dioxins over MnOx-CeO2/TiO2 catalysts: A combined experimental and density functional theory study.

Elucidation of the catalytic decomposition mechanism of dioxins is pivotal in developing highly efficient dioxin degradation catalysts. In order to accurately simulate the whole molecular structure of dioxins, two model compounds, o-dichlorobenzene (o-DCB) and furan, were employed to represent the chlorinated benzene ring and oxygenated central ring within a dioxin molecule, respectively. Experiments and Density Functional Theory (DFT) calculations were combined to investigate the adsorption as well as oxidation of o-DCB and furan over MnOx-CeO2/TiO2 catalyst (denoted as MnCe/Ti). The results indicate that competitive adsorption exists between furan and o-DCB. The former exhibits superior adsorption capacity on MnCe/Ti catalyst at 100 °C - 150 °C, for it can adsorb on both surface metal atom and surface oxygen vacancies (Ov) via its O-terminal; while the latter adsorbs primarily by anchoring its Cl atom to surface Ov. Regarding oxidation, furan can be completely oxidized at 150 °C - 300 °C with a high CO2 selectivity (above 80 %). However, o-DCB cannot be totally oxidized and the resulting intermediates cause the deactivation of catalyst. Interestingly, the pre-adsorption of furan on catalyst surface can facilitate the catalytic oxidation of o-DCB below 200 °C, possibly because the dissociated adsorption of furan may form additional reactive oxygen species on catalyst surface. Therefore, this work provides new insights into the catalytic decomposition mechanism of dioxins as well as the optimization strategies for developing dioxin-degradation catalysts with high efficiency at low temperature.

Transcriptome and Metabolome Analyses Reflect the Molecular Mechanism of Drought Tolerance in Sweet Potato.

Sweet potato (Ipomoea batatas (L.) Lam.) is one of the most widely cultivated crops in the world, with outstanding stress tolerance, but drought stress can lead to a significant decrease in its yield. To reveal the response mechanism of sweet potato to drought stress, an integrated physiological, transcriptome and metabolome investigations were conducted in the leaves of two sweet potato varieties, drought-tolerant zhenghong23 (Z23) and a more sensitive variety, jinong432 (J432). The results for the physiological indexes of drought showed that the peroxidase (POD) and superoxide dismutase (SOD) activities of Z23 were 3.68 and 1.21 times higher than those of J432 under severe drought, while Z23 had a higher antioxidant capacity. Transcriptome and metabolome analysis showed the importance of the amino acid metabolism, respiratory metabolism, and antioxidant systems in drought tolerance. In Z23, amino acids such as asparagine participated in energy production during drought by providing substrates for the citrate cycle (TCA cycle) and glycolysis (EMP). A stronger respiratory metabolism ability could better maintain the energy supply level under drought stress. Drought stress also activated the expression of the genes encoding to antioxidant enzymes and the biosynthesis of flavonoids such as rutin, resulting in improved tolerance to drought. This study provides new insights into the molecular mechanisms of drought tolerance in sweet potato.

Stalled translation by mitochondrial stress upregulates a CNOT4-ZNF598 ribosomal quality control pathway important for tissue homeostasis.

Translational control exerts immediate effect on the composition, abundance, and integrity of the proteome. Ribosome-associated quality control (RQC) handles ribosomes stalled at the elongation and termination steps of translation, with ZNF598 in mammals and Hel2 in yeast serving as key sensors of translation stalling and coordinators of downstream resolution of collided ribosomes, termination of stalled translation, and removal of faulty translation products. The physiological regulation of RQC in general and ZNF598 in particular in multicellular settings is underexplored. Here we show that ZNF598 undergoes regulatory K63-linked ubiquitination in a CNOT4-dependent manner and is upregulated upon mitochondrial stresses in mammalian cells and Drosophila. ZNF598 promotes resolution of stalled ribosomes and protects against mitochondrial stress in a ubiquitination-dependent fashion. In Drosophila models of neurodegenerative diseases and patient cells, ZNF598 overexpression aborts stalled translation of mitochondrial outer membrane-associated mRNAs, removes faulty translation products causal of disease, and improves mitochondrial and tissue health. These results shed lights on the regulation of ZNF598 and its functional role in mitochondrial and tissue homeostasis.

Microfluidics-Prepared Ultra-small Biomimetic Nanovesicles for Brain Tumor Targeting.

Blood-brain-barrier (BBB) serves as a fatal guard of the central nervous system as well as a formidable obstacle for the treatment of brain diseases such as brain tumors. Cell membrane-derived nanomedicines are promising drug carriers to achieve BBB-penetrating and brain lesion targeting. However, the challenge of precise size control of such nanomedicines has severely limited their therapeutic effect and clinical application in brain diseases. To address this problem, this work develops a microfluidic mixing platform that enables the fabrication of cell membrane-derived nanovesicles with precise controllability and tunability in particle size and component. Sub-100 nm macrophage plasma membrane-derived vesicles as small as 51 nm (nanoscale macrophage vesicles, NMVs), with a narrow size distribution (polydispersity index, PDI: 0.27) and a high drug loading rate (up to 89% for indocyanine green-loaded NMVs, NMVs@ICG (ICG is indocyanine green)), are achieved through a one-step process. Compared to beyond-100 nm macrophage cell membrane vesicles (general macrophage vesicles, GMVs) prepared via the traditional methods, the new NMVs exhibits rapid (within 1 h post-injection) and enhanced orthotopic glioma targeting (up to 78% enhancement), with no extra surface modification. This work demonstrates the great potential of such real-nanoscale cell membrane-derived nanomedicines in targeted brain tumor theranostics.

Improved encapsulation effect and structural properties of whey protein isolate by dielectric barrier discharge cold plasma.

The whey protein isolate (WPI) was modified by dielectric barrier discharge cold plasma (DBD) in order to improve its encapsulation efficiency of rutin. In this work, the effect of DBD treatment on structure and physicochemical properties of WPI and the interaction between DBD-treated WPI and rutin were investigated. The results showed that the structural change of WPI leaded to the exposure of internal hydrophobic groups, increasing the interaction site with rutin. The encapsulation efficiency of DBD-treated WPI (30 kV, 30 s) on rutin was improved by 12.42 % compared with control group. The results of multispectral analysis showed that static quenching occurred in the process of interaction between DBD-treated and rutin, hydrogen bond and van der Waals force were the main forces between them. Therefore, DBD treatment can be used as a method to improve the encapsulation efficiency of WPI on hydrophobic active substances.

Revisiting multi-view learning: A perspective of implicitly heterogeneous Graph Convolutional Network.

Graph Convolutional Network (GCN) has become a hotspot in graph-based machine learning due to its powerful graph processing capability. Most of the existing GCN-based approaches are designed for single-view data. In numerous practical scenarios, data is expressed through multiple views, rather than a single view. The ability of GCN to model homogeneous graphs is indisputable, while it is insufficient in facing the heterophily property of multi-view data. In this paper, we revisit multi-view learning to propose an implicit heterogeneous graph convolutional network that efficiently captures the heterogeneity of multi-view data while exploiting the powerful feature aggregation capability of GCN. We automatically assign optimal importance to each view when constructing the meta-path graph. High-order cross-view meta-paths are explored based on the obtained graph, and a series of graph matrices are generated. Combining graph matrices with learnable global feature representation to obtain heterogeneous graph embeddings at various levels. Finally, in order to effectively utilize both local and global information, we introduce a graph-level attention mechanism at the meta-path level that allocates private information to each node individually. Extensive experimental results convincingly support the superior performance of the proposed method compared to other state-of-the-art approaches.

Functional and Transcriptomic Characterization of Postnatal Maturation of ENS and SIP Syncytium in Mice Colon.

The interplay of the enteric nervous system (ENS) and SIP syncytium (smooth muscle cells-interstitial cells of Cajal-PDGFRα+ cells) plays an important role in the regulation of gastrointestinal (GI) motility. This study aimed to investigate the dynamic regulatory mechanisms of the ENS-SIP system on colon motility during postnatal development. Colonic samples of postnatal 1-week-old (PW1), 3-week-old (PW3), and 5-week-old (PW5) mice were characterized by RNA sequencing, qPCR, Western blotting, isometric force recordings (IFR), and colonic motor complex (CMC) force measurements. Our study showed that the transcriptional expression of Pdgfrα, c-Kit, P2ry1, Nos1, and Slc18a3, and the protein expression of nNOS, c-Kit, and ANO1 significantly increased with age from PW1 to PW5. In PW1 and PW3 mice, colonic migrating movement was not fully developed. In PW5 mice, rhythmic CMCs were recorded, similar to the CMC pattern described previously in adult mice. The inhibition of nNOS revealed excitatory and non-propulsive responses which are normally suppressed due to ongoing nitrergic inhibition. During postnatal development, molecular data demonstrated the establishment and expansion of ICC and PDGFRα+ cells, along with nitrergic and cholinergic nerves and purinergic receptors. Our findings are important for understanding the role of the SIP syncytium in generating and establishing CMCs in postnatal, developing murine colons.

Design and Synthesis of Novel Near-Infrared Fluorescence Probes Based on an Open Conformation of a Cytochrome P450 1B1 Complex for Molecular Imaging of Colorectal Tumors.

Cytochrome P450 1B1 (CYP1B1) is induced during the early stage of cancer and is universally overexpressed in tumors. Thus, it was considered as a potential biomarker for the monitoring of cancer. In this study, we designed and synthesized CYP1B1-targeted near-infrared (NIR) fluorescence molecular probes based on the latest reported open conformation of the CYP1B1-α-naphthoflavone (ANF) complex. According to the architecture of the open channel, we introduced linkers and a Cy5.5 fragment at the 5' position of ANF derivatives with strong CYP1B1 inhibitory activity to obtain probes 19-21. Then, in vitro cell-based studies showed that the probes could be enriched in tumor cells by binding to CYP1B1. During in vivo and ex vivo imaging in a xenograft mouse model, probe 19 with the best binding affinity was proven to be able to identify tumor sites in both fluorescence imaging and photoacoustic imaging modes.

Foregroundness-Aware Task Disentanglement and Self-Paced Curriculum Learning for Domain Adaptive Object Detection.

Unsupervised domain adaptive object detection (UDA-OD) is a challenging problem since it needs to locate and recognize objects while maintaining the generalization ability across domains. Most existing UDA-OD methods directly integrate the adaptive modules into the detectors. This integration procedure can significantly sacrifice the detection performances, though it enhances the generalization ability. To solve this problem, we propose an effective framework, named foregroundness-aware task disentanglement and self-paced curriculum adaptation (FA-TDCA), to disentangle the UDA-OD task into four independent subtasks of source detector pretraining, classification adaptation, location adaptation, and target detector training. The disentanglement can transfer the knowledge effectively while maintaining the detection performance of our model. In addition, we propose a new metric, i.e., foregroundness, and use it to evaluate the confidence of the location result. We use both foregroundness and classification confidence to assess the label quality of the proposals. For effective knowledge transfer across domains, we utilize a self-paced curriculum learning paradigm to train adaptors and gradually improve the quality of the pseudolabels associated with the target samples. Experiment results indicate that our method achieves state-of-the-art results on four cross-domain object detection tasks.

Age-dependent seasonal growth cessation in Populus.

In temperate and boreal regions, perennial plants adapt their annual growth cycle to the change of seasons. In natural forests, juvenile seedlings usually display longer growth seasons compared to adult trees to ensure their establishment and survival under canopy shade. However, how trees adjust their annual growth according to their age is not known. In this study, we show that age-dependent seasonal growth cessation is genetically controlled and found that the miR156-SPL3/5 module, a key regulon of vegetative phase change (VPC), also triggers age-dependent growth cessation in Populus trees. We show that miR156 promotes shoot elongation during vegetative growth, and its targets SPL3/5s function in the same pathway but as repressors. We find that the miR156-SPL3/5s regulon controls growth cessation in both leaves and shoot apices and through multiple pathways, but with a different mechanism compared to how the miR156-SPL regulon controls VPC in annual plants. Taken together, our results reveal an age-dependent genetic network in mediating seasonal growth cessation, a key phenological process in the climate adaptation of perennial trees.

Learnable Graph Convolutional Network With Semisupervised Graph Information Bottleneck.

Graph convolutional network (GCN) has gained widespread attention in semisupervised classification tasks. Recent studies show that GCN-based methods have achieved decent performance in numerous fields. However, most of the existing methods generally adopted a fixed graph that cannot dynamically capture both local and global relationships. This is because the hidden and important relationships may not be directed exhibited in the fixed structure, causing the degraded performance of semisupervised classification tasks. Moreover, the missing and noisy data yielded by the fixed graph may result in wrong connections, thereby disturbing the representation learning process. To cope with these issues, this article proposes a learnable GCN-based framework, aiming to obtain the optimal graph structures by jointly integrating graph learning and feature propagation in a unified network. Besides, to capture the optimal graph representations, this article designs dual-GCN-based meta-channels to simultaneously explore local and global relations during the training process. To minimize the interference of the noisy data, a semisupervised graph information bottleneck (SGIB) is introduced to conduct the graph structural learning (GSL) for acquiring the minimal sufficient representations. Concretely, SGIB aims to maximize the mutual information of both the same and different meta-channels by designing the constraints between them, thereby improving the node classification performance in the downstream tasks. Extensive experimental results on real-world datasets demonstrate the robustness of the proposed model, which outperforms state-of-the-art methods with fixed-structure graphs.

Joint learning of feature and topology for multi-view graph convolutional network.

Graph convolutional network has been extensively employed in semi-supervised classification tasks. Although some studies have attempted to leverage graph convolutional networks to explore multi-view data, they mostly consider the fusion of feature and topology individually, leading to the underutilization of the consistency and complementarity of multi-view data. In this paper, we propose an end-to-end joint fusion framework that aims to simultaneously conduct a consistent feature integration and an adaptive topology adjustment. Specifically, to capture the feature consistency, we construct a deep matrix decomposition module, which maps data from different views onto a feature space obtaining a consistent feature representation. Moreover, we design a more flexible graph convolution that allows to adaptively learn a more robust topology. A dynamic topology can greatly reduce the influence of unreliable information, which acquires a more adaptive representation. As a result, our method jointly designs an effective feature fusion module and a topology adjustment module, and lets these two modules mutually enhance each other. It takes full advantage of the consistency and complementarity to better capture the more intrinsic information. The experimental results indicate that our method surpasses state-of-the-art semi-supervised classification methods.