scCompressSA: dual-channel self-attention based deep autoencoder model for single-cell clustering by compressing gene-gene interactions.

BACKGROUND: Single-cell clustering has played an important role in exploring the molecular mechanisms about cell differentiation and human diseases. Due to highly-stochastic transcriptomics data, accurate detection of cell types is still challenged, especially for RNA-sequencing data from human beings. In this case, deep neural networks have been increasingly employed to mine cell type specific patterns and have outperformed statistic approaches in cell clustering. RESULTS: Using cross-correlation to capture gene-gene interactions, this study proposes the scCompressSA method to integrate topological patterns from scRNA-seq data, with support of self-attention (SA) based coefficient compression (CC) block. This SA-based CC block is able to extract and employ static gene-gene interactions from scRNA-seq data. This proposed scCompressSA method has enhanced clustering accuracy in multiple benchmark scRNA-seq datasets by integrating topological and temporal features. CONCLUSION: Static gene-gene interactions have been extracted as temporal features to boost clustering performance in single-cell clustering For the scCompressSA method, dual-channel SA based CC block is able to integrate topological features and has exhibited extraordinary detection accuracy compared with previous clustering approaches that only employ temporal patterns.

Time-gated single-pixel imaging of Cherenkov emission from a medical linear accelerator.

Cherenkov imaging is an ideal tool for real-time in vivo verification of a radiation therapy dose. Given that radiation is pulsed from a medical linear accelerator (LINAC) together with weak Cherenkov emissions, time-gated high-sensitivity imaging is required for robust measurements. Instead of using an expensive camera system with limited efficiency of detection in each pixel, a single-pixel imaging (SPI) approach that maintains promising sensitivity over the entire spectral band could be used to provide a low-cost and viable alternative. A prototype SPI system was developed and demonstrated here in Cherenkov imaging of LINAC dose delivery to a water tank. Validation experiments were performed using four regular fields and an intensity-modulated radiotherapy (IMRT) delivery plan. The Cherenkov image-based projection percent depth dose curves (pPDDs) were compared to pPDDs simulated by the treatment planning system (TPS), with an overall average error of 0.48, 0.42, 0.65, and 1.08% for the 3, 5, 7, and 9 cm square beams, respectively. The composite image of the IMRT plan achieved a 85.9% pass rate using 3%/3 mm gamma index criteria, in comparing Cherenkov intensity and TPS dose. This study validates the feasibility of applying SPI to the Cherenkov imaging of radiotherapy dose for the first time to our knowledge.

SSA Net: Small Scale-Aware Enhancement Network for Human Pose Estimation.

In the field of human pose estimation, heatmap-based methods have emerged as the dominant approach, and numerous studies have achieved remarkable performance based on this technique. However, the inherent drawbacks of heatmaps lead to serious performance degradation in methods based on heatmaps for smaller-scale persons. While some researchers have attempted to tackle this issue by improving the performance of small-scale persons, their efforts have been hampered by the continued reliance on heatmap-based methods. To address this issue, this paper proposes the SSA Net, which aims to enhance the detection accuracy of small-scale persons as much as possible while maintaining a balanced perception of persons at other scales. SSA Net utilizes HRNetW48 as a feature extractor and leverages the TDAA module to enhance small-scale perception. Furthermore, it abandons heatmap-based methods and instead adopts coordinate vector regression to represent keypoints. Notably, SSA Net achieved an AP of 77.4% on the COCO Validation dataset, which is superior to other heatmap-based methods. Additionally, it achieved highly competitive results on the Tiny Validation and MPII datasets as well.

Compressed sensing based dynamic MR image reconstruction by using 3D-total generalized variation and tensor decomposition: k-t TGV-TD.

PURPOSE: Compressed Sensing Magnetic Resonance Imaging (CS-MRI) is a promising technique to accelerate dynamic cardiac MR imaging (DCMRI). For DCMRI, the CS-MRI usually exploits image signal sparsity and low-rank property to reconstruct dynamic images from the undersampled k-space data. In this paper, a novel CS algorithm is investigated to improve dynamic cardiac MR image reconstruction quality under the condition of minimizing the k-space recording. METHODS: The sparse representation of 3D cardiac magnetic resonance data is implemented by synergistically integrating 3D total generalized variation (3D-TGV) algorithm and high order singular value decomposition (HOSVD) based Tensor Decomposition, termed k-t TGV-TD method. In the proposed method, the low rank structure of the 3D dynamic cardiac MR data is performed with the HOSVD method, and the localized image sparsity is achieved by the 3D-TGV method. Moreover, the Fast Composite Splitting Algorithm (FCSA) method, combining the variable splitting with operator splitting techniques, is employed to solve the low-rank and sparse problem. Two different cardiac MR datasets (cardiac perfusion and cine MR datasets) are used to evaluate the performance of the proposed method. RESULTS: Compared with the state-of-art methods, such as k-t SLR, 3D-TGV, HOSVD based tensor decomposition and low-rank plus sparse method, the proposed k-t TGV-TD method can offer improved reconstruction accuracy in terms of higher peak SNR (PSNR) and structural similarity index (SSIM). The proposed k-t TGV-TD method can achieve significantly better and stable reconstruction results than state-of-the-art methods in terms of both PSNR and SSIM, especially for cardiac perfusion MR dataset. CONCLUSIONS: This work proved that the k-t TGV-TD method was an effective sparse representation way for DCMRI, which was capable of significantly improving the reconstruction accuracy with different acceleration factors.

Transcriptome profiling unveils GAP43 regulates ABC transporters and EIF2 signaling in colorectal cancer cells.

BACKGROUND: The growth- and plasticity-associated protein-43 (GAP43) is biasedly expressed in indigestive system and nervous system. Recent study has shown that GAP43 is responsible for the development of neuronal growth and axonal regeneration in normal nervous tissue, while serves as a specific biomarker of relapsed or refractory neuroblastoma. However, its expression pattern and function in digestive system cancer remains to be clarified. METHODS: In this study, we examined the GAP43 status with qRT-PCR and bisulfite genomic sequencing in colorectal cancer (CRC). We investigated the effect of overexpressed GAP43 in CRC cells with RNA-seq. The RNA-seq data was analyzed with DAVID and IPA. RESULTS: GAP43 was downregulated in CRC compared to the adjacent tissues. DNA methylase inhibitor 5-Aza-CdR treatment could significantly induce GAP43, indicated that the silencing of GAP43 gene in CRC is closely related to DNA methylation. Bisulfite genomic sequencing confirmed the promoter methylation of GAP43 in CRC. To explore the transcriptional alterations by overexpressed GAP43 in CRC, we performed RNA-seq and found that upregulated genes were significantly enriched in the signaling pathways of ABC transporters and ECM-receptor interaction, while downregulated genes were significantly enriched in Ribosome signaling pathway. Further Ingenuity Pathway Analysis (IPA) showed that EIF2 signaling pathway was significantly repressed by overexpression of GAP43. CONCLUSION: Our findings provide a novel mechanistic insight of GAP43 in CRC. Transcriptome profiling of overexpressed GAP43 in CRC uncovered the functional roles of GAP43 in the development of human CRC.

CYP2C9 inhibits the invasion and migration of esophageal squamous cell carcinoma via downregulation of HDAC.

Cytochrome P450 2C9 (CYP2C9) is involved in the metabolism of cancer drugs and exogenous carcinogens. In our study, CYP2C9 was downregulated in multiple cohorts of human esophageal squamous cell carcinoma (ESCC). Until now, its role and epigenetic regulation of CYP2C9 repression in ESCC remain poorly understood. CYP2C9 repression in collected ESCC patient tumor tissues was demonstrated by RT-qPCR and Western blot. The histone acetylation level was carried out by the treatment of histone deacetylase inhibitor TSA and RNA interference. Epigenetic analysis revealed that the increased expression of CYP2C9 in KYSE-150 and TE1 cells was characterized by inhibition of HDAC8 and HDAC1, respectively. TSA decreased the levels of HDAC occupancy around CYP2C9 promoter region greatly. Overexpression of CYP2C9 reduced the invasion and migration of ESCC cells.

Upregulation of OATP1A2 in human oesophageal squamous cell carcinoma cells via the HDAC6-GCN5/PCAF-H3K9Ac axis.

1. OATP1A2 overexpressed is involved in chemotherapy disposition, indicating its role in tumour development and progression.2. CHIP and siRNA were used to evaluate the status of histone acetylation at the OATP1A2 promoter. The role of OATP1A2 was analysed by gene-set enrichment and overall survival analysis.3. OATP1A2 expression levels in ESCC was notably higher than that in para-cancer tissues. OATP1A2 high expression are associated with bile salt metabolic pathway and poor prognosis. Furthermore, HDAC6 was repressed in ESCC, increasing the levels of H3K9Ac catalysed by GCN5/PCAF at the OATP1A2 promoter region.4. Abnormal histone hyperacetylation mediated by the HDAC6-GCN5/PCAF-H3K9Ac axis resulted in increased OATP1A2 expression in ESCC, and OATP1A2 may serve as a promising prognostic biomarker for ESCC.5. In conclusion, this study indicated that suppression of OATP1A2 would inhibit the progression and prognosis in ESCC.

Insight into unique somitogenesis of yak (Bos grunniens) with one additional thoracic vertebra.

BACKGROUND: The yak is a species of livestock which is crucial for local communities of the Qinghai-Tibet Plateau and adjacent regions and naturally owns one more thoracic vertebra than cattle. Recently, a sub-population of yak termed as the Jinchuan yak has been identified with over half its members own a thoracolumbar vertebral formula of T15L5 instead of the natural T14L5 arrangement. The novel T15L5 positioning is a preferred genetic trait leading to enhanced meat and milk production. Selective breeding of this trait would have great agricultural value and exploration of the molecular mechanisms underlying this trait would both accelerate this process and provide us insight into the development and regulation of somitogenesis. RESULTS: Here we investigated the genetic background of the Jinchuan yak through resequencing fifteen individuals, comprising five T15L5 individuals and ten T14L5 individuals with an average sequencing depth of > 10X, whose thoracolumbar vertebral formulae were confirmed by anatomical observation. Principal component analysis, linkage disequilibrium analysis, phylogenetic analysis, and selective sweep analysis were carried out to explore Jinchuan yak's genetic background. Three hundred and thirty candidate markers were identified as associated with the additional thoracic vertebrae and target sequencing was used to validate seven carefully selected markers in an additional 51 Jinchuan yaks. The accuracies of predicting 15 thoracic vertebrae and 20 thoracolumbar vertebrae with these 7 markers were 100.00 and 33.33% despite they both could only represent 20% of all possible genetic diversity. Two genes, PPP2R2B and TBLR1, were found to harbour the most candidate markers associated with the trait and likely contribute to the unique somitic number and identity according to their reported roles in the mechanism of somitogenesis. CONCLUSIONS: Our findings provide a clear depiction of the Jinchuan yak's genetic background and a solid foundation for marker-assistant selection. Further exploitation of this unique population and trait could be promoted with the aid of our genomic resource.

FAM60A promotes cisplatin resistance in lung cancer cells by activating SKP2 expression.

Cisplatin is a widely used chemotherapeutic drug in lung cancer treatment. Most cancer patients eventually develop cisplatin resistance, resulting in a poor prognosis. Previously, we identified a novel marker, family with sequence similarity 60A (FAM60A), that was responsible for resistance in cisplatin-resistant human lung adenocarcinoma A549 (A549/DDP) cells. Here, we investigated the biological effects of FAM60A in A549/DDP cells and explored the underlying molecular mechanisms to understand its functional role in cisplatin resistance. Real-time quantitative PCR and western blot analysis were used to determine the expression levels of FAM60A in A549/DDP cells. FAM60A and SKP2 were knockdown with small-interfering RNA (siRNA). Cancer cell viability was analyzed with flow cytometry. The mRNA and protein expression levels of FAM60A increased significantly and dose-dependently in A549/DDP cells following cisplatin treatment. FAM60A overexpression up-regulated MDR1 expression, inhibited caspase 3, cleaved-caspase 3, and caspase 8 expression, and prevented cancer cell death. Microarray analysis of cells transfected with siRNA against the FAM60A transcript and control samples showed that SKP2 expression was positively regulated by FAM60A. SKP2 knockdown using a short-hairpin RNA reversed the functions induced by FAM60A. These results suggest that overexpression of FAM60A in A549/DDP cells led to SKP2 upregulation and enhanced cisplatin resistance in cancer cells. These provide new insights into chemoresistance and may contribute to reversing cisplatin resistance during lung cancer treatment.

Identification and verification of differentially expressed genes in yak mammary tissue during the lactation cycle.

Yaks (Bos grunniens) live primarily in the Qinghai-Tibetan plateau (altitude: 2000-5000 m). Their milk presents unusual characteristics, containing large amounts of solids including fat and protein, and it is, therefore, important to understand the genetic makeup of the yak. To identify potentially critical genes playing a role in yak mammary tissue from colostrum to mature milk phase of lactogenesis, the early lactation (colostrum) stage (ELS; day 1 after parturition) and mature lactation (milk) stage (MLS; day 15) were chosen for comparison. An ELS-specific cDNA library was established by suppression subtractive hybridization and 25 expressed sequence tags at ELS were identified by sequencing and alignment. To further confirm our results the expression levels of 21 genes during the lactation cycle were measured using quantitative real-time RT-PCR (qRT-PCR). The qRT-PCR results confirmed 9 significantly up-regulated genes at ELS vs. MLS in yak mammary tissue, in which the l-amino acid oxidase 1 (LAO1) and collagen, type I, alpha I (COL1A1) were the most significantly up-regulated. During the lactation cycle, the highest expression of some milk fat genes (i.e., XDH and FABP3) in yak mammary tissue appears earlier than that in dairy cow. Our data also indicate MYC potentially playing a central role through putative regulation of COL1A1, CD44, SPARC, FASN and GPAM.

[Research on electrocardiogram classification using deep residual network with pyramid convolution structure].

Recently, deep neural networks (DNNs) have been widely used in the field of electrocardiogram (ECG) signal classification, but the previous models have limited ability to extract features from raw ECG data. In this paper, a deep residual network model based on pyramidal convolutional layers (PC-DRN) was proposed to implement ECG signal classification. The pyramidal convolutional (PC) layer could simultaneously extract multi-scale features from the original ECG data. And then, a deep residual network was designed to train the classification model for arrhythmia detection. The public dataset provided by the physionet computing in cardiology challenge 2017(CinC2017) was used to validate the classification experiment of 4 types of ECG data. In this paper, the harmonic mean F 1 of classification accuracy and recall was selected as the evaluation indexes. The experimental results showed that the average sequence level F 1 ( SeqF 1) of PC-DRN was improved from 0.857 to 0.920, and the average set level F 1 ( SetF 1) was improved from 0.876 to 0.925. Therefore, the PC-DRN model proposed in this paper provided a promising way for the feature extraction and classification of ECG signals, and provided an effective tool for arrhythmia classification.

Synthetic promoter for efficient and muscle-specific expression of exogenous genes.

Synthetic promoters (SPs) have many advantages over their natural counterparts, especially with regard to transcriptional activity and tissue specificity. Here, we report a new strategy to construct SPs for efficient and muscle-specific gene expression. First, 19 nucleic acid motifs classified to 3 kinds of transcriptional regulatory elements were rationally selected. A recombinant promoter library was constructed by randomly assembling these motifs. Second, the transcriptional activities of ~1200 SPs were screened by intramuscular expression of several reporter genes in different cell lines for activity higher than that of the cytomegalovirus (CMV) promoter, with SP-301 finally identified as the strongest. A single intramuscular injection of mice with an SP-301 plasmid expressing mouse growth hormone releasing hormone accelerated mouse growth significantly over 24 days. Third, the muscle specificity of SP-301 was confirmed in transgenic mice. Finally, in comparison with the CMV promoter, SP-301 accelerated translocation and increased the level of plasmid in the nuclei of myoblast cells to a greater extent than in non-muscle cells. Altogether, the study has provided a more rational strategy to construct efficient and tissue-specific promoters, with the promoter SP-301 exhibiting promising potential for establishing an intramuscular gene expression system for therapeutic applications.

Multiplex Assay for Identifying Animal Species Found in the Tibetan Area Using the Mitochondrial 12S rRNA Gene.

Southwestern China has an area with unique natural conditions located in alpine regions at altitudes from 2000 to 5000 m; this area is referred to as the Qinghai-Tibetan plateau (QTP). Unique animals, such as yaks (Bos grunniens), are found extensively on the plateau of Southwestern China due to its unique environment. In recent years, the prevalence of fake meat products such as fake jerky has increased in this area. This research was conducted as an attempt to develop a reliable multiplex polymerase chain reaction (mPCR) detection method for identifying nine animal species found in QTP. We developed the mPCR method using the specific sites found in 12S rRNA region of these nine species, which was effective in discriminating between the nine species and was successful in terms of validated reproducibility, detection limit (<6 pg total DNA), discrimination of mixed samples, and specificity (approximately 99%) using real meat samples. Our results show that the mPCR detection method can overcome the limitations of prior detection methods, such as restriction fragment length polymorphism or high-resolution melting analysis methods.

Non-rigid structure estimation in trajectory space from monocular vision.

In this paper, the problem of non-rigid structure estimation in trajectory space from monocular vision is investigated. Similar to the Point Trajectory Approach (PTA), based on characteristic points' trajectories described by a predefined Discrete Cosine Transform (DCT) basis, the structure matrix was also calculated by using a factorization method. To further optimize the non-rigid structure estimation from monocular vision, the rank minimization problem about structure matrix is proposed to implement the non-rigid structure estimation by introducing the basic low-rank condition. Moreover, the Accelerated Proximal Gradient (APG) algorithm is proposed to solve the rank minimization problem, and the initial structure matrix calculated by the PTA method is optimized. The APG algorithm can converge to efficient solutions quickly and lessen the reconstruction error obviously. The reconstruction results of real image sequences indicate that the proposed approach runs reliably, and effectively improves the accuracy of non-rigid structure estimation from monocular vision.

Expression, Purification, and Characteristic of Tibetan Sheep Breast Lysozyme Using Pichia pastoris Expression System.

A lysozyme gene from breast of Tibetan sheep was successfully expressed by secretion using a-factor signal sequence in the methylotrophic yeast, Pichia pastoris GS115. An expression yield and specific activity greater than 500 mg/L and 4,000 U/mg was obtained. Results at optimal pH and temperature showed recombinant lysozyme has higher lytic activity at pH 6.5 and 45°C. This study demonstrates the successful expression of recombinant lysozyme using the eukaryotic host organism P. pastoris paving the way for protein engineering. Additionally, this study shows the feasibility of subsequent industrial manufacture of the enzyme with this expression system together with a high purity scheme for easy high-yield purification.

Metabolic profiling of triptophenolide in rat plasma, urine, bile and faeces after intragastric administration.

Triptophenolide, a major diterpenoid extracted from Tripterygium wilfordii Hook. f., has been reported to possess significant anti-tumour, anti-androgen and anti-inflammatory activities. However, the metabolic fate of triptophenolide remains unknown. Therefore, this study focused on the metabolic profiling of triptophenolide in rat plasma, urine, bile and faeces following intragastric administration. An ultraperformance liquid chromatography coupled to quadrupole time-of-flight mass spectrometry with combination of extracted ion chromatogram strategy based on 71 typical metabolic reactions was established to comprehensively profile the metabolites of triptophenolide. This strategy allowed for the identification of 17 metabolites from the biosamples. Reduction, oxidation, glucuronide conjugation, and hydroxylation were considered as its main metabolic pathways in vivo. The present study will be greatly helpful for the further pharmacological studies on triptophenolide and would provide valuable information for its clinical application.

Arrhythmia classification based on multi-feature multi-path parallel deep convolutional neural networks and improved focal loss.

Early diagnosis of abnormal electrocardiogram (ECG) signals can provide useful information for the prevention and detection of arrhythmia diseases. Due to the similarities in Normal beat (N) and Supraventricular Premature Beat (S) categories and imbalance of ECG categories, arrhythmia classification cannot achieve satisfactory classification results under the inter-patient assessment paradigm. In this paper, a multi-path parallel deep convolutional neural network was proposed for arrhythmia classification. Furthermore, a global average RR interval was introduced to address the issue of similarities between N vs. S categories, and a weighted loss function was developed to solve the imbalance problem using the dynamically adjusted weights based on the proportion of each class in the input batch. The MIT-BIH arrhythmia dataset was used to validate the classification performances of the proposed method. Experimental results under the intra-patient evaluation paradigm and inter-patient evaluation paradigm showed that the proposed method could achieve better classification results than other methods. Among them, the accuracy, average sensitivity, average precision, and average specificity under the intra-patient paradigm were 98.73%, 94.89%, 89.38%, and 98.24%, respectively. The accuracy, average sensitivity, average precision, and average specificity under the inter-patient paradigm were 91.22%, 89.91%, 68.23%, and 95.23%, respectively.

Myocardial infarction detection method based on the continuous T-wave area feature and multi-lead-fusion deep features.

Objective.Myocardial infarction (MI) is one of the most threatening cardiovascular diseases. This paper aims to explore a method for using an algorithm to autonomously classify MI based on the electrocardiogram (ECG).Approach.A detection method of MI that fuses continuous T-wave area (C_TWA) feature and ECG deep features is proposed. This method consists of three main parts: (1) The onset of MI is often accompanied by changes in the shape of the T-wave in the ECG, thus the area of the T-wave displayed on different heartbeats will be quite different. The adaptive sliding window method is used to detect the start and end of the T-wave, and calculate the C_TWA on the same ECG record. Additionally, the coefficient of variation of C_TWA is defined as the C_TWA feature of the ECG. (2) The multi lead fusion convolutional neural network was implemented to extract the deep features of the ECG. (3) The C_TWA feature and deep features of the ECG were fused by soft attention, and then inputted into the multi-layer perceptron to obtain the detection result.Main results.According to the inter-patient paradigm, the proposed method reached a 97.67% accuracy, 96.59% precision, and 98.96% recall on the PTB dataset, as well as reached 93.15% accuracy, 93.20% precision, and 95.14% recall on the clinical dataset.Significance.This method accurately extracts the feature of the C_TWA, and combines the deep features of the signal, thereby improving the detection accuracy and achieving favorable results on clinical datasets.

LUCMT: Learnable under-sampling and reconstructed network with cross multi-head attention transformer for accelerating MR image reconstruction.

BACKGROUND AND OBJECTIVE: As a widely used technique for Magnetic Resonance Image (MRI) acceleration, compressed sensing MRI involves two main issues: designing an effective sampling strategy and reconstructing the image from significantly under-sampled K-space data. In this paper, an innovative approach is proposed to address these two challenges simultaneously. METHODS: A novel MRI reconstruction method, termed as LUCMT, is implemented by integrating a learnable under-sampling strategy with a reconstruction network based on the Cross Multi-head Attention Transformer. In contrast to conventional static sampling methods, the proposed adaptive sampling scheme is processed optimally by learning the optimal sampling technique, which involves binarizing the sampling pattern by a sigmoid function and computing gradients by backpropagation. And the reconstruction network is designed by using CS-MRI depth unfolding network that incorporates a Cross Multi-head Attention (CMA) module with inertial and gradient descent terms. RESULTS: T1 brain MR images from the FastMRI dataset are used to validate the performance of the proposed method. A series of experiments are conducted to validate the superior performance of our proposed network in terms of quantitative metrics and visual quality. Compared with other state-of-the-art reconstruction methods, LUCMT achieves better reconstruction performances with more accurate details. Specifically, LUCMT achieves PSNR and SSIM results of 41.87/0.9749, 46.64/0.9868, 50.41/0.9924, and 53.51/0.9955 at sampling rates of 10 %, 20 %, 30 %, and 40 %, respectively. CONCLUSIONS: The proposed LUCMT method can provide a promising way for generating optimal under-sampling mask and accelerating MRI reconstruction accurately.

Data governance and Gensini score automatic calculation for coronary angiography with deep-learning-based natural language extraction.

With the widespread adoption of electronic health records, the amount of stored medical data has been increasing. Clinical data, often in the form of semi-structured or unstructured electronic medical records (EMRs), contains rich patient information. However, due to the use of natural language by physicians when composing these records, the effectiveness of traditional methods such as dictionaries, rule matching, and machine learning in the extraction of information from these unstructured texts falls short of clinical standards. In this paper, a novel deep-learning-based natural language extraction method is proposed to overcome current shortcomings in data governance and Gensini score automatic calculation in coronary angiography. A pre-trained model called bidirectional encoder representation from transformers (BERT) with strong text feature representation capabilities is employed as the feature representation layer. It is combined with bidirectional long short-term memory (BiLSTM) and conditional random field (CRF) models to extract both global and local features from the text. The study included an evaluation of the model on a dataset from a hospital in China and it was compared with another model to validate its practical advantages. Hence, the BiLSTM-CRF model was employed to automatically extract relevant coronary angiogram information from EMR texts. The achieved F1 score was 91.19, which is approximately 0.87 higher than the BERT-BiLSTM-CRF model.