Hypothyroidism is one of the common endocrine diseases, and its incidence is increasing year by year. Due to the insidious nature of this disease, it often leads to delayed treatment and even misdiagnosis. This paper proposes ILSHIP, an interpretable predictive model for hypothyroidism, to reduce its diagnostic complexity as well as improve the predictive performance and interpretability of existing models. First, the ILSHIP prediction model was built based on label encoding, missing value processing, feature selection, and data enhancement of the dataset. Second, the comprehensive performance of ILSHIP was compared with twelve existing related study models and eleven mainstream models, such as XGBoost and MLP. The experimental results showed that, based on the optimal hyperparameters the ILSHIP model can achieve 99.392%, 99.437%, 99.348%, 99.381%, and 99.960% in accuracy, recall, specificity, F1, and AUC, respectively. The accuracy of the ILSHIP model was about 0.7%-15.4% higher than the existing models. By introducing the SHAP framework into the ILSHIP model, important features affecting hypothyroidism such as thyroid stimulating hormone (TSH) and free thyroxine index (FTI) were also identified, and the influencing factors for different individuals were finally analyzed to provide a basis for medical personnel to monitor the condition.

BACKGROUND: Deformable image registration is crucial for multiple radiation therapy applications. Fast registration of computed tomography (CT) lung images is challenging because of the large and nonlinear deformation between inspiration and expiration. With advancements in deep learning techniques, learning-based registration methods are considered efficient alternatives to traditional methods in terms of accuracy and computational cost. METHOD: In this study, an unsupervised lung registration network (LRN) with cycle-consistent training is proposed to align two acquired CT-derived lung datasets during breath-holds at inspiratory and expiratory levels without utilizing any ground-truth registration results. Generally, the LRN model uses three loss functions: image similarity, regularization, and Jacobian determinant. Here, LRN was trained on the CT datasets of 705 subjects and tested using 10 pairs of public CT DIR-Lab datasets. Furthermore, to evaluate the effectiveness of the registration technique, target registration errors (TREs) of the LRN model were compared with those of the conventional algorithm (sum of squared tissue volume difference; SSTVD) and a state-of-the-art unsupervised registration method (VoxelMorph). RESULTS: The results showed that the LRN with an average TRE of 1.78 ± 1.56 mm outperformed VoxelMorph with an average TRE of 2.43 ± 2.43 mm, which is comparable to that of SSTVD with an average TRE of 1.66 ± 1.49 mm. In addition, estimating the displacement vector field without any folding voxel consumed less than 2 s, demonstrating the superiority of the learning-based method with respect to fiducial marker tracking and the overall soft tissue alignment with a nearly real-time speed. CONCLUSIONS: Therefore, this proposed method shows significant potential for use in time-sensitive pulmonary studies, such as lung motion tracking and image-guided surgery.

Parameter estimation of neuronal networks is closely related with information processing mechanisms in neural systems. Estimation of synaptic parameters for neuronal networks was an time consuming task. Due to complex interactions between neurons, computational efficiency and accuracy of estimation methods is relatively low. Meanwhile, inherent topological properties such as core-periphery and modular structures are not fully considered in estimation. In order to improve the efficiency and accuracy of estimation, this study proposes a two-stage PartitionMLE method which introduces detected neuronal modules as topological constraints in estimation. The proposed PartitionMLE method firstly decomposes the system into multiple non-overlapping neuronal modules, by performing topology-based module detection. Dynamic parameters including intra-modular and inter-modular parameters are estimated in two stages, using detected hubs to connect non-overlapping neuronal modules. The contributions of PartitionMLE method are two-folds: reducing estimation errors and improving the model interpretability. Experiments about neuronal networks consisting of Hodgkin-Huxley (HH) and leaky integrate-and-firing (LIF) neurons validated the effectiveness of the PartitionMLE method, with comparison to the single-stage MLE method.

Chronic obstructive pulmonary disease is a kind of chronic lung disease characterized by persistent air flow obstruction, which was the third leading cause of death in China. The incidence of COPD is steadily and increasing and has been a globally sever disease. Accordingly, it is urgently needed to explore how to diagnose and treat COPD timely. This study aims to find key genes to diagnose COPD as soon as possible to avoid COPD processing and analyze immune cell infiltration between COPD early stage and late stage. Two GEO datasets were merged as the merge data for analyses. 157 DEGs were used for GSEA analysis to find the pathway between COPD early stage and late stage. Above all, gene EXPH5 stood out from the screen as the most likely candidate diagnosis biomarker of COPD indicating the late-stage by least LASSO and SVM-RFE. ROC curves of EXPH5 were applied to represent the discriminatory ability through the area under the curve which is the gold standard to evaluate the accuracy of diagnosis and survival rate. The CIBERSORT algorithm was used to assess the distribution of tissue-infiltrating immune cells between two COPD stages. The diagnosis biomarker, gene EXPH5 had a positive correlation with NK cells resting; mast cell resting, eosinophils, and negative correlation with T cell gamma delta, macrophages M1, which underscore the role of gene and immune cell infiltration. To make results more reliable, we further analyzed the gene EXPH5 expression in single-cell transcriptome data and showed again that EXPH5 genes significantly downregulated in the late stage of COPD especially in the main lung cell types AT1 and AT2. In a word, our study identified genes EXPH5 as a marker gene, which adds to the knowledge for clinical diagnosis and pharmaceutical design of COPD.

Herein we have made a comparative study of the efficiency of three different nanotubes viz. Carbon nanotube (CNT), boron nitride nanotube (BNNT) and silicon carbide nanotube (SiCNT) to deliver the cancerous drug, Azacitidine (AZD). The atomistic description of the encapsulation process of AZD in these nanotubes has been analyzed by evaluating parameters like adsorption energy, electrostatic potential map, reduced density gradient (RDG). Higher adsorption energy of AZD with BNNT (-0.66eV), SiCNT (-0.92eV) compared to CNT (-0.56eV) confirms stronger binding affinity of the drug for the former than the later. Charge density and electrostatic potential map suggest that charge separation involving BNNT and CNT is more prominent than SiCNT. Evaluation of different thermodynamic parameters like Gibbs free energy, enthalpy change revealed that the overall encapsulation process is spontaneous and exothermic in nature and much favorable with BNNT and SiCNT. Stabilizing interactions of the drug with BNNT and SiCNT has been confirmed from RDG analysis. ADMP molecular dynamics simulation supports that the encapsulation process of the drug within the NT at room temperature. These results open up unlimited opportunities for the applications of these NTs as a drug delivery system in the field of nanomedicine.

To solve the problems of high latency, high system overhead, and small supported scale in the current application of pharmaceutical traceability combined with blockchain technology, an algorithm called Pharmaceutical-Practical Byzantine Fault Tolerance (P-PBFT) based on PBFT, grouping, and credit voting is proposed. The algorithm combines the characteristics of a pharmaceutical supply chain, optimizes the consistency protocol in the original algorithm, divides large-scale network nodes into different consensus sets by response speed, and performs grouping consensus. The algorithm's credit model and voting mechanism dynamically updates user status according to the behavior of nodes in consensus, evaluates the reliability of users, and also serves as a basis for electing management nodes. Experimental results show that the improved P-PBFT consensus algorithm provides smaller latency and higher throughput for pharmaceutical traceability systems, supports larger-scale traceability, effectively alleviates the dramatic increase in communication among network nodes, and reduces the influence of malicious nodes.

The evolution of MDR and XDR-TB is a growing concern and public health safety threat around the world. Gene mutations are the prime cause of drug resistance in tuberculosis, however the reports of double mutations further aggravated the situation. Despite the large-scale genomic sequencing and identification of novel mutations, structure investigation of the protein is still required to structurally and functionally characterize these novel mutations to design novel drugs for improved clinical outcome. Hence, we used structural bioinformatics approaches i.e. molecular modeling, residues communication and molecular simulation to understand the impact of novel double S59Y-L85P, D86G-V180F and S104G-V130 M mutation on the structure, function of pncA encoded Pyrazinamidase (PZase) and resistance of Pyrazinamide (PZA). Our results revealed that these mutations alter the binding paradigm and destabilize the protein to release the drug. Protein commination network (PCN) revealed variations in the hub residues and sub-networks which consequently alter the internal communication and signaling. The region 1-75 demonstrated higher flexibility in the mutant structures and minimal by the wild type which destabilize of the internally arranged beta-sheets which consequently reduce the binding of PZA and potentially Fe ion in the mutants. Hydrogen bonding analysis further validated the findings. The total binding free energy (ΔG) for each complex i.e. wild type -7.46 kcal/mol, S59Y-L85P -5.21 kcal/mol, S104G-V130 M -5.33 kcal/mol while for the D86G-V180F mutant the TBE was calculated to be -6.26 kcal/mol. This further confirms that these mutations reduce the binding energy of PZA for PZase and causes resistance in the effective therapy for TB. The trajectories motion was also observed to be affected by these mutations. In conclusion, these mutations use destabilizing approach to reduce the binding of PZA and causes resistance. These features can be used to design novel structure-based drugs against Tuberculosis.

State-of-the-art (SOTA) convolutional neural network models have been widely adapted in medical imaging and applied to address different clinical problems. However, the complexity and scale of such models may not be justified in medical imaging and subject to the available resource budget. Further increasing the number of representative feature maps for the classification task decreases the model explainability. The current data normalization practice is fixed prior to model development and discounting the specification of the data domain. Acknowledging these issues, the current work proposed a new scalable model family called PlexusNet; the block architecture and model scaling by the network's depth, width, and branch regulate PlexusNet's architecture. The efficient computation costs outlined the dimensions of PlexusNet scaling and design. PlexusNet includes a new learnable data normalization algorithm for better data generalization. We applied a simple yet effective neural architecture search to design PlexusNet tailored to five clinical classification problems that achieve a performance noninferior to the SOTA models ResNet-18 and EfficientNet B0/1. It also does so with lower parameter capacity and representative feature maps in ten-fold ranges than the smallest SOTA models with comparable performance. The visualization of representative features revealed distinguishable clusters associated with categories based on latent features generated by PlexusNet. The package and source code are at https://github.com/oeminaga/PlexusNet.git.

Cells are the basic units of biological organization, and the quantitative analysis of cellular states is an important topic in medicine and is valuable in revealing the complex mechanisms of microscopic world organisms. In order to better understand cell cycle changes as well as drug actions, we need to track cell migration and division. In this paper, we propose a novel engineering model for tracking cells using cell position and motion fields (CPMF). The training sample does not need to be manually annotated, and we modify and edit it against the ground truth using auxiliary tools. The core idea of the project is to combine detection and correlation, and the cell sequence samples are trained by a U-Net network model composed of 3D CNNs, which can track the migration, division, and entry and exit of cells in the field of view with high accuracy in all directions. The average detection accuracy of the cell coordinates is 98.38% and the average tracking accuracy is 98.70%.

The computer-aided diagnosis system based on dermoscopic images has played an important role in the clinical treatment of skin lesion. An accurate, efficient, and automatic skin lesion segmentation method is an important auxiliary tool for clinical diagnosis. At present, skin lesion segmentation still suffers from great challenges. Existing deep-learning-based automatic segmentation methods frequently use convolutional neural networks (CNN). However, the globally-sharing feature re-weighting vector may not be optimal for the prediction of lesion areas in dermoscopic images. The presence of hairs and spots in some samples aggravates the interference of similar categories, and reduces the segmentation accuracy. To solve this problem, this paper proposes a new deep network for precise skin lesion segmentation based on a U-shape structure. To be specific, two lightweight attention modules: adaptive channel-context-aware pyramid attention (ACCAPA) module and global feature fusion (GFF) module, are embedded in the network. The ACCAPA module can model the characteristics of the lesion areas by dynamically learning the channel information, contextual information and global structure information. GFF is used for different levels of semantic information interaction between encoder and decoder layers. To validate the effectiveness of the proposed method, we test the performance of ACCPG-Net on several public skin lesion datasets. The results show that our method achieves better segmentation performance compared to other state-of-the-art methods.

PURPOSE: To develop a general unsupervised anomaly detection method based only on MR images of normal brains to automatically detect various brain abnormalities. MATERIALS AND METHODS: In this study, a novel method based on three-dimensional deep autoencoder network is proposed to automatically detect and segment various brain abnormalities without being trained on any abnormal samples. A total of 578 normal T2w MR volumes without obvious abnormalities were used for model training and validation. The proposed 3D autoencoder was evaluated on two different datasets (BraTs dataset and in-house dataset) containing T2w volumes from patients with glioblastoma, multiple sclerosis and cerebral infarction. Lesions detection and segmentation performance were reported as AUC, precision-recall curve, sensitivity, and Dice score. RESULTS: In anomaly detection, AUCs for three typical lesions were as follows: glioblastoma, 0.844; multiple sclerosis, 0.858; cerebral infarction, 0.807. In anomaly segmentation, the mean Dice for glioblastomas was 0.462. The proposed network also has the ability to generate an anomaly heatmap for visualization purpose. CONCLUSION: Our proposed method was able to automatically detect various brain anomalies such as glioblastoma, multiple sclerosis, and cerebral infarction. This work suggests that unsupervised anomaly detection is a powerful approach to detect arbitrary brain abnormalities without labeled samples. It has the potential to support diagnostic workflow in radiology as an automated tool for computer-aided image analysis.

OBJECTIVE: The aim of this study is to propose a method to reduce the sensitivity of the estimated omnipolar electrogram (oEGM) with respect to the angle of the propagation wavefront. METHODS: A novel configuration of cliques taking into account all four electrodes of a squared cell is proposed. To test this approach, simulations of HD grids of cardiac activations at different propagation angles, conduction velocities, interelectrode distance and electrogram waveforms are considered. RESULTS: The proposed approach successfully provided narrower loops (essentially a straight line) of the electrical field described by the bipole pair with respect to the conventional approach. Estimation of the direction of propagation was improved. Additionally, estimated oEGMs presented larger amplitude, and estimations of the local activation times were more accurate. CONCLUSIONS: A novel method to improve the estimation of oEGMs in HD grid of electrodes is proposed. This approach is superior to the existing methods and avoids pitfalls not yet resolved. RELEVANCE: Robust tools for quantifying the cardiac substrate are crucial to determine with accuracy target ablation sites during an electrophysiological procedure.

Specialized microbial communities in the fungus-farming termite gut and fungal comb microbiome help maintain host nutrition through interactive biochemical activities of complex carbohydrate degradation. Numerous research studies have been focused on identifying the microbial species in the termite gut and fungal comb microbiota, but the community-wide metabolic interaction patterns remain obscure. The inter-microbial metabolic interactions in the community environment are essential for executing biochemical processes like complex carbohydrate degradation and maintaining the host's physicochemical homeostasis. Recent progress in high-throughput sequencing techniques and mathematical modeling provides suitable platforms for constructing multispecies genome-scale community metabolic models that can render sound knowledge about microbial metabolic interaction patterns. Here, we have implemented the genome-scale metabolic modeling strategy to map the relationship between genes, proteins, and reactions of 12 key bacterial species from fungal cultivating termite gut and fungal comb microbiota. The resulting individual genome-scale metabolic models (GEMs) have been analyzed using flux balance analysis (FBA) to optimize the metabolic flux distribution pattern. Further, these individual GEMs have been integrated into genome-scale community metabolic models where a heuristics-based computational procedure has been employed to track the inter-microbial metabolic interactions. Two separate genome-scale community metabolic models were reconstructed for the O. badius gut and fungal comb microbiome. Analysis of the community models showed up to ∼167% increased flux range in lignocellulose degradation, amino acid biosynthesis, and nucleotide metabolism pathways. The inter-microbial metabolic exchange of amino acids, SCFAs, and small sugars was also upregulated in the multispecies community for maximum biomass formation. The flux variability analysis (FVA) has also been performed to calculate the feasible flux range of metabolic reactions. Furthermore, based on the calculated metabolic flux values, newly defined parameters, i.e., pairwise metabolic assistance (PMA) and community metabolic assistance (CMA) showed that the microbial species are getting up to 15% higher metabolic benefits in the multispecies community compared to pairwise growth. Assessment of the inter-microbial metabolic interaction patterns through pairwise growth support index (PGSI) indicated an increased mutualistic interaction in the termite gut environment compared to the fungal comb. Thus, this genome-scale community modeling study provides a systematic methodology to understand the inter-microbial interaction patterns with several newly defined parameters like PMA, CMA, and PGSI. The microbial metabolic assistance and interaction patterns derived from this computational approach will enhance the understanding of combinatorial microbial activities and may help develop effective synergistic microcosms to utilize complex plant polymers.

Microscopic analysis of breast cancer images is the primary task in diagnosing cancer malignancy. Recent attempts to automate this task have employed deep learning models whose success has depended on large volumes of data, while acquiring annotated data in biomedical domains is time-consuming and may not always be feasible. A typical strategy to address this is to apply transfer learning using pre-trained models on a large natural image database (e.g., ImageNet) instead of training a model from scratch. This approach, however, has not been effective in several previous studies due to fundamental differences between natural and medical images. In this study, for the first time we proposed the idea of using a compatible data set of histopathological images to classify breast cancer cytological biopsy specimens. Despite intrinsic differences between histopathological and cytological images, we demonstrate that the features learned by deep networks during the pre-training procedure are compatible with those obtained throughout fine-tuning process. To thoroughly investigate this assertion, we explore three different strategies for training as well as two different approaches for fine-tuning deep learning models. By comparing the obtained results with those of previous state-of-the-art research conducted on the same data set, we demonstrate that the proposed method boasts of improved classification accuracy by 6% to 17% compared to the studies which were based on traditional machine learning techniques, and also enhanced accuracy by roughly 7% compared to those who utilized deep learning methods, eventually achieving 98.73% validation accuracy and 94.55% test accuracy. Exploring different training scenarios also revealed that using a compatible dataset has helped to elevate the classification accuracy by 3.0% compared to the typical approach of using ImageNet. Experimental results show that our approach, despite using a very small number of training images, has achieved performance comparable to that of experienced pathologists and has the potential to be applied in clinical settings.

Vessel segmentation in fundus images is a key procedure in the diagnosis of ophthalmic diseases, which can play a role in assisting doctors in diagnosis. Although current deep learning-based methods can achieve high accuracy in segmenting fundus vessel images, the results are not satisfactory in segmenting microscopic vessels that are close to the background region. The reason for this problem is that thin blood vessels contain very little information, with the convolution operation of each layer in the deep network, this part of the information will be randomly lost. To improve the segmentation ability of the small blood vessel region, a multi-input network (MINet) was proposed to segment vascular regions more accurately. We designed a multi-input fusion module (MIF) in the encoder, which is proposed to acquire multi-scale features in the encoder stage while preserving the microvessel feature information. In addition, to further aggregate multi-scale information from adjacent regions, a multi-scale atrous spatial pyramid (MASP) module is proposed. This module can enhance the extraction of vascular information without reducing the resolution loss. In order to better recover segmentation results with details, we designed a refinement module, which acts on the last layer of the network output to refine the results of the last layer of the network to get more accurate segmentation results. We use the HRF, CHASE_DB1 public dataset to validate the fundus vessel segmentation performance of the MINet model. Also, we merged these two public datasets with our collected Ultra-widefield fundus image (UWF) data as one dataset to test the generalization ability of the model. Experimental results show that MINet achieves an F1 score of 0.8324 on the microvessel segmentation task, achieving a high accuracy compared to the current mainstream models.

Acral melanoma (AM), a rare subtype of cutaneous melanoma, shows higher incidence in Asians, including Koreans, than in Caucasians. However, the genetic modification associated with AM in Koreans is not well known and has not been comprehensively investigated in terms of oncogenic signaling, and hallmarks of cancer. We performed whole-exome and RNA sequencing for Korean patients with AM and acquired the genetic alterations and gene expression profiles. KIT alterations (previously known to be recurrent alterations in AM) and CDK4/CCND1 copy number amplifications were identified in the patients. Genetic and transcriptomic alterations in patients with AM were functionally converge to the hallmarks of cancer and oncogenic pathways, including 'proliferative signal persistence', 'apoptotic resistance', and 'activation of invasion and metastasis', despite the heterogeneous somatic mutation profiles of Korean patients with AM. This study may provide a molecular understanding for therapeutic strategy for AM.

Electrical brain stimulation is a treatment method for brain disorder patients. The majority of patients with a severe brain disorder have brain atrophy. However, it is not clearly understood if electrical brain stimulation is effective even to brain atrophy. In this work, we developed anatomical head models with varying degrees of brain atrophy, so that we could investigate the effects of subdural/epidural cortical stimulations. The correlation between brain atrophy and cortical stimulation was quantified by calculating the effective volume that cortical stimulation influenced in this brain atrophy simulation study. The results showed that the effective volumes in both cortical stimulations decreased significantly with brain atrophy. There was also a strong correlation (0.9989) between the cerebrospinal fluid (CSF) and brain atrophy. The increase in CSF volume following brain atrophy reinforced the shunting effect between the brain and CSF and appeared to be the cause of a decrease in the stimulation effect on the brain. Overall, the epidural cortical stimulation was more sensitive (up to 57%) to the severity of the brain atrophy than the subdural cortical stimulation.

Computer-aided lung cancer diagnosis (CAD) system on computed tomography (CT) helps radiologists guide preoperative planning and prognosis assessment. The flexibility and scalability of deep learning methods are limited in lung CAD. In essence, two significant challenges to be solved are (1) Label scarcity due to cost annotations of CT images by experienced domain experts, and (2) Label inconsistency between the observed nodule malignancy and the patients' pathology evaluation. These two issues can be considered weak label problems. We address these issues in this paper by introducing a weakly-supervised lung cancer detection and diagnosis network (WS-LungNet), consisting of a semi-supervised computer-aided detection (Semi-CADe) that can segment 3D pulmonary nodules based on unlabeled data through adversarial learning to reduce label scarcity, as well as a cross-nodule attention computer-aided diagnosis (CNA-CADx) for evaluating malignancy at the patient level by modeling correlations between nodules via cross-attention mechanisms and thereby eliminating label inconsistency. Through extensive evaluations on the LIDC-IDRI public database, we show that our proposed method achieves 82.99% competition performance metric (CPM) on pulmonary nodule detection and 88.63% area under the curve (AUC) on lung cancer diagnosis. Extensive experiments demonstrate the advantage of WS-LungNet on nodule detection and malignancy evaluation tasks. Our promising results demonstrate the benefits and flexibility of the semi-supervised segmentation with adversarial learning and the nodule instance correlation learning with the attention mechanism. The results also suggest that making use of the unlabeled data and taking the relationship among nodules in a case into account are essential for lung cancer detection and diagnosis.

Follicle-stimulating hormone receptor (FSHR) is a glycoprotein hormone receptor that plays a vital role in reproduction, cancer progression and osteoporosis. Owing to its therapeutic importance, several small molecule modulators have been identified by researchers through high throughput studies that usually include virtual screening of chemical libraries followed by in vitro validation through radio-ligand binding assays, cAMP accumulation and luciferase-based luminescence assays. The binding site of these modulators and structural changes that accompany modulator binding remains elusive. Here, we address these aspects through molecular docking and MD simulations on well-studied FSHR modulators and comparing the domain motions between agonist/FSH bound and antagonist bound FSHR structures. It was observed that agonist and antagonist modulators bind to the same site, but interact with distinct residues in transmembrane domain(TMD). FSHR(TMD) residues Ile522, Ala595, Ile602 and Val604 were found to interact only with agonist. Notably, these residues are conserved in the close homolog luteinizing hormone/choriogonadotropin receptor (LHCGR) and participate in interaction with its agonist Org43553. We observed distinctly prominent domain motions and conformational changes in TM helices 3, 4 and 6 for agonist bound FSHR structure. These structural changes have also been reported for LHCGR, and few GPCR members suggesting an important and well conserved mechanism of GPHR activation that could be exploited for design of novel modulators.

BACKGROUND: Lung adenocarcinoma (LUAD) remains a global health concern with its poor prognosis and high mortality. Whether tumor cells invade through the basement membrane (BM) is the key factor to determine the prognosis of LUAD. This study aimed to identify the BM-related gene signatures to improve the overall prognosis of LUAD. MATERIALS & METHODS: A series of bioinformatics analyses were conducted based on TCGA and GEO datasets. Unsupervised consistent cluster analysis was performed, and 500 LUAD patients were assigned to two different groups according to expressions of 222 BM-related genes. The differentially expressed genes (DEGs) between the two clusters were identified, and Lasso regression, ROC curve, univariate and multivariate Cox regression analyses and enrichment analysis were conducted. Besides, ssGSEA, CIBERSORT and ESTIMATE algorithmwere were employed to understand the relationship between the tumor microenvironment (TME) and risk scores. Moreover, single cell clustering and trajectory analyses were performed to further understand the significance of BM-related genes. Finally, qRT-PCR was used to verify the prognosis model. RESULTS: A total of 31 prognostic BM-related genes were determined for LUAD, and a novel 17-mRNA prognostic model named BMsocre was successfully established to predict the overall survival of LUAD patients. The high BMscore group indicated worse prognosis. Seventeen DEGs were enriched mainly in metabolism, ECM-receptor interaction and immune response. In addition, the high-risk group showed higher TMB and lower immune score. The low-risk group had a better immunotherapeutic response where immune escape was less likely. The BMscore model was verified in our patient cohort. Furthermore, NELL2 was mainly expressed in clusters of T cells, and was identified to play a critical role in T-cell differentiation. CONCLUSIONS: A novel BMscore model was successfully established and might be effective for providing guidance to LUAD therapy.