MultiGran-SMILES: multi-granularity SMILES learning for molecular property prediction.

MOTIVATION: Extracting useful molecular features is essential for molecular property prediction. Atom-level representation is a common representation of molecules, ignoring the sub-structure or branch information of molecules to some extent; however, it is vice versa for the substring-level representation. Both atom-level and substring-level representations may lose the neighborhood or spatial information of molecules. While molecular graph representation aggregating the neighborhood information of a molecule has a weak ability in expressing the chiral molecules or symmetrical structure. In this article, we aim to make use of the advantages of representations in different granularities simultaneously for molecular property prediction. To this end, we propose a fusion model named MultiGran-SMILES, which integrates the molecular features of atoms, sub-structures and graphs from the input. Compared with the single granularity representation of molecules, our method leverages the advantages of various granularity representations simultaneously and adjusts the contribution of each type of representation adaptively for molecular property prediction. RESULTS: The experimental results show that our MultiGran-SMILES method achieves state-of-the-art performance on BBBP, LogP, HIV and ClinTox datasets. For the BACE, FDA and Tox21 datasets, the results are comparable with the state-of-the-art models. Moreover, the experimental results show that the gains of our proposed method are bigger for the molecules with obvious functional groups or branches. AVAILABILITY AND IMPLEMENTATION: The code and data underlying this work are available on GitHub at https://github. com/Jiangjing0122/MultiGran. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

TranGRU: focusing on both the local and global information of molecules for molecular property prediction.

Molecular property prediction is an essential but challenging task in drug discovery. The recurrent neural network (RNN) and Transformer are the mainstream methods for sequence modeling, and both have been successfully applied independently for molecular property prediction. As the local information and global information of molecules are very important for molecular properties, we aim to integrate the bi-directional gated recurrent unit (BiGRU) into the original Transformer encoder, together with self-attention to better capture local and global molecular information simultaneously. To this end, we propose the TranGRU approach, which encodes the local and global information of molecules by using the BiGRU and self-attention, respectively. Then, we use a gate mechanism to reasonably fuse the two molecular representations. In this way, we enhance the ability of the proposed model to encode both local and global molecular information. Compared to the baselines and state-of-the-art methods when treating each task as a single-task classification on Tox21, the proposed approach outperforms the baselines on 9 out of 12 tasks and state-of-the-art methods on 5 out of 12 tasks. TranGRU also obtains the best ROC-AUC scores on BBBP, FDA, LogP, and Tox21 (multitask classification) and has a comparable performance on ToxCast, BACE, and ecoli. On the whole, TranGRU achieves better performance for molecular property prediction. The source code is available in GitHub: https://github.com/Jiangjing0122/TranGRU.

A seed coat-specific ß-ketoacyl-CoA synthase, KCS12, is critical for preserving seed physical dormancy.

Physical dormancy in seeds exists widely in seed plants and plays a vital role in maintaining natural seed banks. The outermost cuticle of the seed coat forms a water-impermeable layer, which is critical for establishing seed physical dormancy. We previously set up the legume plant Medicago truncatula as an excellent model for studying seed physical dormancy, and our studies revealed that a class II KNOTTED-like homeobox, KNOX4, is a transcription factor critical for controlling hardseededness. Here we report the function of a seed coat ß-ketoacyl-CoA synthase, KCS12. The expression level of KCS12 is significantly downregulated in the knox4 mutant. The KCS12 gene is predominantly expressed in the seed coat, and seed development in the M. truncatula kcs12 mutant is altered. Further investigation demonstrated that kcs12 mutant seeds lost physical dormancy and were able to absorb water without scarification treatment. Chemical analysis revealed that concentrations of C24:0 lipid polyester monomers are significantly decreased in mutant seeds, indicating that KCS12 is an enzyme that controls the production of very long chain lipid species in the seed coat. A chromatin immunoprecipitation assay demonstrated that the expression of KCS12 in the seed coat is directly regulated by the KNOX4 transcription factor. These findings define a molecular mechanism by which KNOX4 and KCS12 control formation of the seed coat and seed physical dormancy.

Identification of a gene responsible for seedpod spine formation and other phenotypic alterations using whole-genome sequencing analysis in Medicago truncatula.

In nature, some plant species produce seedpods with spines, which is an adaptive biological trait for protecting the seed and helping seed dispersal. However, the molecular mechanism of spine formation is still unclear. While conducting routine tissue culture and transformation in the model legume Medicago truncatula, we identified a smooth seedpod (ssp1) mutant with a suite of other phenotypic changes. Preliminary analysis showed that the mutation was derived from the tissue culture process. Genetic segregation analysis suggested that ssp1 is a recessive mutant. By combining whole-genome sequencing and bioinformatics analysis, we found that the mutant phenotype was caused by a single nucleotide polymorphism and a 30 bp deletion in the gene locus Medtr4g039430, named SSP1. Complementation of the M. truncatula ssp1 and Arabidopsis twd1 mutants showed complete restoration, indicating that SSP1 is an ortholog of Arabidopsis TWD1 which encodes an immunophilin-like FK506-binding protein 42. The formation of spines on seedpods is associated with auxin transport. The method used in this study offers an effective way for detecting genes responsible for somaclonal variations. The results demonstrate, for the first time, that SSP1 plays a crucial role in the determination of spine formation on seedpods.

The Relationship between Blood Groups and Aortic Dissection.

BACKGROUND: Plenty of diseases have been found having associations with blood types, especially cardiovascular diseases. The purpose of this study was to clarify whether there is a relationship between blood groups and acute aortic dissection. We also further studied the distribution of blood groups in different types of acute aortic dissection. METHODS: A total of 291 patients diagnosed with acute aortic dissection from 2011 to 2018 were enrolled and analyzed retrospectively in this study. The control group consisted of 582 patients who received plastic surgery at West China hospital from 2011 to 2018. First, we analyzed the distribution of blood groups between the study group and the control group, including the ABO, Rh, O and non-O groups. Then, we further divided the study group into two groups by the type of acute aortic dissection to determine if there was difference in blood groups between the two types of acute aortic dissection. RESULTS: The analysis of the distribution of ABO blood groups (p = 0.302) and Rh blood groups (p = 0.502) did not reveal statistically significant differences. There were no statistically significant differences in the distributions of ABO blood groups and Rh blood groups in different types of acute aortic dissection. CONCLUSIONS: Our study did not prove the incidence of acute aortic dissection, or the type of acute aortic dissection had a relationship with common blood groups.

Double Thread Suspension: A Novel Technique to Facilitate End-to-Side Venous Anastomosis With a Microvascular Coupling Device in Head and Neck Reconstruction.

PURPOSE: Microvascular free tissue transfer has become a mainstay for head and neck reconstruction, and vascular anastomosis is 1 of the most challenging aspects. This study aimed to explore a safe, convenient, and efficient technique for end-to-side venous anastomosis using the coupling device. MATERIALS AND METHODS: We evaluated a novel surgical strategy, double thread suspension (DTS), by reviewing data collected in a consecutive series of 60 patients who underwent free tissue grafting after surgical resection of head and neck tumors. Patients who underwent end-to-side microvascular anastomosis using the microvascular coupler device were randomly divided into the DTS group (n = 30) and the traditional procedure group (TRA, n = 30), the demographic data, coupler size, anastomosis time, microvascular complications, and flap survival were determined. RESULTS: The optional size range of coupler devices in the DTS group is better than that of the conventional group, ranging from 1.5 to 4.0 mm compared with 2.0 to 4.0 mm. The average anastomosis time was 4.68 ± 0.43 minutes in the DTS group which was significantly lower than the conventional group that was 9.24 ± 1.46 minutes (P < .0001) . There was no statistically significant difference between the 2 groups in coupler related complications or flap survival. CONCLUSION: The novel DTS technique provides the advantages of lowering procedure complexity, reducing operative time, and preventing tearing of vessel wall, therefore making DTS-guided end-to-side anastomosis a safe, relatively straightforward, and reliable technique.

Effect of Preoperative Oral Carbohydrate Administration on Patients Undergoing Cesarean Section with Epidural Anesthesia: A Pilot Study.

PURPOSE: The aim of this study was to evaluate the effect of preoperative oral carbohydrate administration on patients undergoing Cesarean section with epidural anesthesia. DESIGN: Randomized controlled clinical study. METHODS: A total of 75 patients undergoing Cesarean section (American Society of Anesthesiologists physical status grade I-II) were randomized to preparation with a carbohydrate drink (CHO group), flavored water (placebo group), or to the fasting group. The CHO and placebo groups were double-blinded and given 300 mL of the drink 2 hours before surgery. Visual analog scores of the patient were assessed to evaluate thirst, hunger, and anxiety level, and the gastric antral cross-sectional areas were recorded by ultrasonography during the operative period. Insulin resistance was calculated on the basis of the blood glucose and insulin levels assessed before administration and after surgery. FINDINGS: The CHO and placebo groups did not show an increase in gastric fluid volumes in terms of gastric antral cross-sectional area, and there were no adverse events. The visual analog scale scores at preoperative baseline were not different between groups . During the preoperative waiting period, preparation with CHO reduced not only thirst and anxiety more efficiently than water (placebo) but also hunger (P < .05), whereas water did not. No difference was observed in insulin resistance between groups before intake of the drink. Compared with the preoperative levels, insulin resistance showed a statistically significant increase in all groups (P < .05); however, the increase was significantly higher in the fasting and placebo groups than in the CHO group (P < .05). CONCLUSIONS: Preoperative administration of CHO decreases postoperative insulin resistance and enhances pregnant women's comfort, leading to a reduced sense of thirst, hunger, and anxiety during the preoperative period for Cesarean section.

LPA-MNI: An Improved Label Propagation Algorithm Based on Modularity and Node Importance for Community Detection.

Community detection is of great significance in understanding the structure of the network. Label propagation algorithm (LPA) is a classical and effective method, but it has the problems of randomness and instability. An improved label propagation algorithm named LPA-MNI is proposed in this study by combining the modularity function and node importance with the original LPA. LPA-MNI first identify the initial communities according to the value of modularity. Subsequently, the label propagation is used to cluster the remaining nodes that have not been assigned to initial communities. Meanwhile, node importance is used to improve the node order of label updating and the mechanism of label selecting when multiple labels are contained by the maximum number of nodes. Extensive experiments are performed on twelve real-world networks and eight groups of synthetic networks, and the results show that LPA-MNI has better accuracy, higher modularity, and more reasonable community numbers when compared with other six algorithms. In addition, LPA-MNI is shown to be more robust than the traditional LPA algorithm.

De novo hydroponics system efficiency for the cuttings of alfalfa (Medicago sativa L.).

The legume plant alfalfa (Medicago sativa L.) is a widely cultivated perennial forage due to its high protein content, palatability, and strong adaptability to diverse agro-ecological zones. Alfalfa is a self-incompatible cross-pollinated autotetraploid species with tetrasomic inheritance. Therefore, maintaining excellent traits through seed reproduction is a prime challenge in alfalfa. However, the cutting propagation technology could enable consistent multiplication of quality plants that are genetically identical to the parent plant. The current study aimed to develop a simple, cost-effective, reproducible, and efficient hydroponic cutting method to preserve alfalfa plants and for molecular research. In this study, alfalfa landrace 'Wudi' was grown in hydroponics for 30 days and used as source material for cuttings. The top, middle and bottom sections of its stem were used as cuttings. The rooting rate, root length, and stem height of the different stem sections were compared to determine the best segment for alfalfa propagation in four nutrient treatments (HM, HM + 1/500H, HM + 1/1000H and d HM + 1/2000H). After 21 days of culture, the rooting rates of all the three stem types under four cutting nutrient solutions were above 78%. The rooting rate of the middle and bottom parts in HM + 1/1000 H and HM + 1/2000 H nutrient solutions reached more than 93%, with a higher health survey score (> 4.70). In conclusion, this study developed a de novo cutting propagation method that can be used to conserve and propagate germplasm in breeding programs and research. This method is a new report on the cutting propagation of alfalfa by hydroponics, which could supplement the existing cutting propagation methods.

[Transcriptome analysis and validation of key genes involved in biosynthesis of iridoids in Gentiana lhassica].

As the main chemical constituents, iridoids are widely distributed within Gentiana, Gentianaceae, with promising bioactivities. Based on the previous work, the transcriptome of G. lhassica, an original plant of Tibetan herb "Jieji Nabao", was sequenced and analyzed in this study, and the transcriptome databases of roots, stems, leaves, and flowers were constructed so as to explore unigenes that may encode the key enzymes in the biosynthetic pathway of iridoids. Then, qRT-PCR was used to validate the relative expression levels of 11 genes named AACT, DXS, MCS, HDS, IDI, GPPS, GES, G10H, 7-DLNGT, 7-DLGT, and SLS in roots, stems, leaves, and flowers. Also, the total contents of gentiopicroside and loganic acid were determined by HPLC, respectively. The results are as follows:(1)a total of 76 486 unigenes with an average length of 852 bp were obtained;(2)335 unigenes were involved in 19 stan-dard secondary metabolism pathways in KEGG database, with phenylpropanoid biosynthesis having the maximum number(75 unigenes), and no isoflavone biosynthetic pathway was annotated;(3)171 unigenes participatedin 27 key enzymes encoding in the biosynthetic pathway of iridoids, and 1-deoxy-D-xylulose-5-phosphate reductoisomerase(DXR) gene was highly expressed;(4)qRT-PCR results were approximately consistent with RNA-Seq data and the relative expression levels of the 11 genes were higher in the aboveground parts(stem, leaf, and flower) than in the underground part(root);(5)the total contents of gentiopicroside and loganic acid were higher in the aboveground parts(stem, leaf, and flower) than in the underground part(root), and the difference was significant. This study provides basic scientific data for accurate species identification, evaluation of germplasm resources, research on secondary pro-duct accumulation of medicinal plants within Gentianaceae, and protection of endangered alpine species.

Multi-task learning models for predicting active compounds.

The computational drug discovery methods can find potential drug-target interactions more efficiently and have been widely studied over past few decades. Such methods explore the relationship between the structural properties of compounds and their biological activity with the assumption that similar compounds tend to share similar biological targets and vice versa. However, traditional Quantitative Structure - Activity Relationship (QSAR) methods often do not have desired accuracy due to insufficient data of compound activity. In this paper, we focus on building Multi-Task Learning (MTL)-based QSAR models by considering multiple similar biological targets together and make shared information transfer across from one task to another, thereby improving not only the learning efficiency, but also the prediction accuracy. This paper selects 6 assay groups with similar biological targets from PubChem and builds their QSAR models with MTL simultaneously. According to the experiment results, our MTL-based QSAR models have better performance over traditional prominent machine learning algorithms and the improvements are even more obvious when other baseline models have low accuracy. The superiority of our models is also proved by Student's t-test with level of significance 5%. Moreover, this paper also explores three different assumptions on the underlying pattern in the dataset and finds that the joint feature MTL models further improve the performance of the QSAR models and are more suitable for building QSAR models for multiple similar biological targets.

γ-Secretase inhibitor reduces immunosuppressive cells and enhances tumour immunity in head and neck squamous cell carcinoma.

Immune evasion is a hallmark feature of cancer, and it plays an important role in tumour initiation and progression. In addition, tumour immune evasion severely hampers the desired antitumour effect in multiple cancers. In this study, we aimed to investigate the role of the Notch pathway in immune evasion in the head and neck squamous cell carcinoma (HNSCC) microenvironment. We first demonstrated that Notch1 signaling was activated in a Tgfbr1/Pten-knockout HNSCC mouse model. Notch signaling inhibition using a γ-secretase inhibitor (GSI-IX, DAPT) decreased tumour burden in the mouse model after prophylactic treatment. In addition, flow cytometry analysis indicated that Notch signaling inhibition reduced the sub-population of myeloid-derived suppressor cells (MDSCs), tumour-associated macrophages (TAMs) and regulatory T cells (Tregs), as well as immune checkpoint molecules (PD1, CTLA4, TIM3 and LAG3), in the circulation and in the tumour. Immunohistochemistry (IHC) of human HNSCC tissues demonstrated that elevation of the Notch1 downstream target HES1 was correlated with MDSC, TAM and Treg markers and with immune checkpoint molecules. These results suggest that modulating the Notch signaling pathway may decrease MDSCs, TAMs, Tregs and immune checkpoint molecules in HNSCC.

Biomarkers: paving stones on the road towards the personalized precision medicine for oral squamous cell carcinoma.

Traditional therapeutics have encountered a bottleneck caused by diagnosis delay and subjective and unreliable assessment. Biomarkers can overcome this bottleneck and guide us toward personalized precision medicine for oral squamous cell carcinoma. To achieve this, it is important to efficiently and accurately screen out specific biomarkers from among the huge number of molecules. Progress in omics-based high-throughput technology has laid a solid foundation for biomarker discovery. With credible and systemic biomarker models, more precise and personalized diagnosis and assessment would be achieved and patients would be more likely to be cured and have a higher quality of life. However, this is not straightforward owing to the complexity of molecules involved in tumorigenesis. In this context, there is a need to focus on tumor heterogeneity and homogeneity, which are discussed in detail. In this review, we aim to provide an understanding of biomarker discovery and application for precision medicine of oral squamous cell carcinoma, and have a strong belief that biomarker will pave the road toward future precision medicine.

Robust MST-Based Clustering Algorithm.

Minimax similarity stresses the connectedness of points via mediating elements rather than favoring high mutual similarity. The grouping principle yields superior clustering results when mining arbitrarily-shaped clusters in data. However, it is not robust against noises and outliers in the data. There are two main problems with the grouping principle: first, a single object that is far away from all other objects defines a separate cluster, and second, two connected clusters would be regarded as two parts of one cluster. In order to solve such problems, we propose robust minimum spanning tree (MST)-based clustering algorithm in this letter. First, we separate the connected objects by applying a density-based coarsening phase, resulting in a low-rank matrix in which the element denotes the supernode by combining a set of nodes. Then a greedy method is presented to partition those supernodes through working on the low-rank matrix. Instead of removing the longest edges from MST, our algorithm groups the data set based on the minimax similarity. Finally, the assignment of all data points can be achieved through their corresponding supernodes. Experimental results on many synthetic and real-world data sets show that our algorithm consistently outperforms compared clustering algorithms.

Effective freeform TIR lens designed for LEDs with high angular color uniformity.

Most LED luminaires face the problem of color nonuniformity in illumination fields, which seriously degrades their lighting quality. To address this issue, we proposed a freeform TIR lens, which could be applied to the most widely adopted phosphor-converted LEDs and multicolor LED arrays. Based on light pattern overlap and energy mapping, the novel TIR lens comprised the freeform panel and the collimator. Analysis of angular color uniformity (ACU), based on a phosphor-converted white LED model, proved that the normalized standard deviation (NSD) of the LED integrated with the freeform TIR lens significantly decreased from 0.792 to 0.232, compared with the NSD of the traditional TIR lens. Moreover, the proposed TIR lens was also applied to a multiple LED array for color mixing. Both the theoretical analysis and simulation results demonstrated their capability to enhance the ACU. These results demonstrated the effectiveness of the proposed method to improve the uniformity of the LED sources regardless of single or multiple LEDs. The improved lighting quality guarantees good employment of LEDs, especially in the fields of entertainment and architectural lighting.

Chloroplast genome structures in Gentiana (Gentianaceae), based on three medicinal alpine plants used in Tibetan herbal medicine.

The genus Gentiana is the largest in the Gentianaceae family with ca. 400 species. However, with most species growing on the Qinghai-Tibet plateau, the processes of adaptive evolution and speciation within the genus is not clear. Also, the genomic analyses could provide important information. So far, the complete chloroplast (cp) genome data of the genus are still deficient. As the second and third sequenced members within Gentianaceae, we report the construction of complete cp sequences of Gentiana robusta King ex Hook. f. and Gentiana crassicaulis Duthie ex Burk., and describe a comparative study of three Gentiana cp genomes, including the cp genome of Gentiana straminea Maxim. published previously. These cp genomes are highly conserved in gene size, gene content, and gene order and the rps16 pseudogene with exon2 missing was found common. Three repeat types and five SSR types were investigated, and the number and distribution are similar among the three genomes. Sixteen genome divergent hotspot regions were identified across these cp genomes that could provide potential molecular markers for further phylogenetic studies in Gentiana. The IR/SC boundary organizations in Gentianales cp genomes were compared and three different types of boundaries were observed. Six data partitions of cp genomes in Gentianales were used for phylogenetic analyses and different data partitions were largely congruent with each other. The ML phylogenetic tree was constructed based on the fragments in cp genomes commonly available in 33 species from Lamiids, including 12 species in Gentianales, 1 in Boraginaceae, 10 in Solanales, and 10 in Lamiales. The result strongly supports the position of Boraginaceae (Ehretia acuminata) as the sister of Solanales, with the bootstrap values of 97 %. This study provides a platform for further research into the molecular phylogenetics of species in the order Gentianales (family Gentianaceae) notably in respect of speciation and species identification.

Authentication of Gentiana straminea Maxim. and its substitutes based on chemical profiling of iridoids using liquid chromatography with mass spectrometry.

A liquid chromatography-electrospray ionization-ion-trap mass spectrometry (MSn ) method was established and applied for authentication of Gentiana straminea from the four substitutes, G. tibetica, G. lhassica, G. waltonii and G. robusta, based on chemical profiling of the principal iridoid glucosides aided by a quadrupole time-of-flight mass spectrometry. The fragmentation pathways of the three representative iridoid glucosides, loganic acid, gentiopicroside and sweroside, were investigated by MSn analysis in negative ion mode, which assisted the characterization of analogs detected in the chromatographic profiling of the tested Gentiana species. In total, 25 iridoids were identified or tentatively characterized from G. straminea and four substitutes, in which 7-O-(4''-O-glucosyl)coumaroyl-loganic acid and 7-O-coumaroyl-loganic acid are diagnostic in G. straminea and can serve as the proposed chemical markers to discriminate it from morphologically similar substitutes.

[A strategy for identifying six species of Sect.Cruciata(Gentiana) in Gansu using DNA barcode sequences].

Located in the transition zone between the Qinghai-Tibet Plateau and the Loess Plateau, Gansu province is one of the distribution centers of Sect. Cruciata, Gentiana (Gentianaceae) in China. Six species in the section, G. crassicaulis, G. straminea, G. siphonantha, G. officinalis, G. dahurica and G . macrophylla, are native to Gansu. In this paper, samples of 6 species and Halenia elliptica (outgroup) were collected. Nuclear DNA ITS, chloroplast DNA matK, rbcL, rpoC1, trnL (UAA) intron, psbA-trnH, atpB-rbcL, trnS (GCU)-trnG (UCC), rpl20-rps12 and trnL (UAA)-trnF (GAA) were sequenced from these samples. Based on the sequence analyses, high intragenomic polymorphisms were detected in ITS regions of G. crassicaulis, G. straminea, G. siphonantha, G. officinalis and G. dahurica, and they showed incomplete concerted evolution. A methodological study to identifying such close-related species as G. macrophylla, G. officinalis and G. dahurica was carried out based on the special genotypes. The results showed that 7 cp DNA sequence fragments could be used to identify G. crassicaulis, G. straminea and G. siphonantha. With nr ITS genotype II,III and IV of G. dahurica, the species can be distinguished from the close-related G. officinalis using 12 cloned sequences in a sample (with statistical significance).The cp DNA sequences of G. macrophylla were classified into two genotypes, and with genotype II, the species can be distinguished from the close-related G. officinalis and G. dahurica using 6 test samples each(with statistical significance). Furthermore, DNA barcode sequences were determined for all 6 species in Gansu. Also, the studies provide some basic data for analyses of genetic diversity and identification of Gentiana species.

[DNA-based identification of Gentiana robusta and related species].

The alpine plant Gentiana robusta is an endemic species to the Sino-Himalayan subregion. Also, it is one of the original plants used as traditional Tibetan medicine Jie-Ji. We sequence the nuclear ribosomal internal transcribed spacer (ITS) regions, matK, rbcL, rpoC1, trnL (UAA), psbA-trnH, atpB-rbcL, trnS( GCU)-trnG(UCC), rpl20-rps12, trnL(UAA)-trnF( GAA) fragments of cp DNA in both G. robusta and such relative species as G. straminea, G. crassicaulis and G. waltonii. With Halenia elliptica as the outgroup, molecular systematic analysis reveals that G. robusta is a natural hybrid. G. straminea is the mother of hybrids, but the father is not very clear. In addition, the molecular markers for distinguishing G. robusta from the parental species or closely related species are identified, respectively. Our studies may provide valuable reference for the species identifications of medicinal plants with complex genetic backgrounds.

[Study on Embryonic Development of Four Species of Gentiana (Gentianaceae)].

OBJECTIVE: To study the embryonic development of Gentiana straminea, G. robusta, G. crassicaulis and G. tibetica. METHODS: The seed germination rates, length and width of embryos, starch grains and chloroplasts were observed and analyzed by statistic software. RESULTS: The seed germination rates of the four species were all high. The increase of the length of embryo depended mainly on the increase of the length of hypocotyl. Starch grains were stored in cells and chloroplasts appeared in cotyledons. The process of germination was divided into eight periods. CONCLUSION: The embryo growth characteristics of the four species are recognized, and the results can be used to study the in situ conservation, genetic breeding and cultivation of Sect. Cruciata.