scPROTEIN: a versatile deep graph contrastive learning framework for single-cell proteomics embedding.

Single-cell proteomics sequencing technology sheds light on protein-protein interactions, posttranslational modifications and proteoform dynamics in the cell. However, the uncertainty estimation for peptide quantification, data missingness, batch effects and high noise hinder the analysis of single-cell proteomic data. It is important to solve this set of tangled problems together, but the existing methods tailored for single-cell transcriptomes cannot fully address this task. Here we propose a versatile framework designed for single-cell proteomics data analysis called scPROTEIN, which consists of peptide uncertainty estimation based on a multitask heteroscedastic regression model and cell embedding generation based on graph contrastive learning. scPROTEIN can estimate the uncertainty of peptide quantification, denoise protein data, remove batch effects and encode single-cell proteomic-specific embeddings in a unified framework. We demonstrate that scPROTEIN is efficient for cell clustering, batch correction, cell type annotation, clinical analysis and spatially resolved proteomic data exploration.

Benchmarking spatial clustering methods with spatially resolved transcriptomics data.

Spatial clustering, which shares an analogy with single-cell clustering, has expanded the scope of tissue physiology studies from cell-centroid to structure-centroid with spatially resolved transcriptomics (SRT) data. Computational methods have undergone remarkable development in recent years, but a comprehensive benchmark study is still lacking. Here we present a benchmark study of 13 computational methods on 34 SRT data (7 datasets). The performance was evaluated on the basis of accuracy, spatial continuity, marker genes detection, scalability, and robustness. We found existing methods were complementary in terms of their performance and functionality, and we provide guidance for selecting appropriate methods for given scenarios. On testing additional 22 challenging datasets, we identified challenges in identifying noncontinuous spatial domains and limitations of existing methods, highlighting their inadequacies in handling recent large-scale tasks. Furthermore, with 145 simulated data, we examined the robustness of these methods against four different factors, and assessed the impact of pre- and postprocessing approaches. Our study offers a comprehensive evaluation of existing spatial clustering methods with SRT data, paving the way for future advancements in this rapidly evolving field.

SODB facilitates comprehensive exploration of spatial omics data.

Spatial omics technologies generate wealthy but highly complex datasets. Here we present Spatial Omics DataBase (SODB), a web-based platform providing both rich data resources and a suite of interactive data analytical modules. SODB currently maintains >2,400 experiments from >25 spatial omics technologies, which are freely accessible as a unified data format compatible with various computational packages. SODB also provides multiple interactive data analytical modules, especially a unique module, Spatial Omics View (SOView). We conduct comprehensive statistical analyses and illustrate the utility of both basic and advanced analytical modules using multiple spatial omics datasets. We demonstrate SOView utility with brain spatial transcriptomics data and recover known anatomical structures. We further delineate functional tissue domains with associated marker genes that were obscured when analyzed using previous methods. We finally show how SODB may efficiently facilitate computational method development. The SODB website is https://gene.ai.tencent.com/SpatialOmics/ . The command-line package is available at https://pysodb.readthedocs.io/en/latest/ .

Senataxin Mutation Reveals How R-Loops Promote Transcription by Blocking DNA Methylation at Gene Promoters.

R-loops are three-stranded nucleic acid structures found abundantly and yet often viewed as by-products of transcription. Studying cells from patients with a motor neuron disease (amyotrophic lateral sclerosis 4 [ALS4]) caused by a mutation in senataxin, we uncovered how R-loops promote transcription. In ALS4 patients, the senataxin mutation depletes R-loops with a consequent effect on gene expression. With fewer R-loops in ALS4 cells, the expression of BAMBI, a negative regulator of transforming growth factor ß (TGF-ß), is reduced; that then leads to the activation of the TGF-ß pathway. We uncovered that genome-wide R-loops influence promoter methylation of over 1,200 human genes. DNA methyl-transferase 1 favors binding to double-stranded DNA over R-loops. Thus, in forming R-loops, nascent RNA blocks DNA methylation and promotes further transcription. Hence, our results show that nucleic acid structures, in addition to sequences, influence the binding and activity of regulatory proteins.

SPDB: a comprehensive resource and knowledgebase for proteomic data at the single-cell resolution.

The single-cell proteomics enables the direct quantification of protein abundance at the single-cell resolution, providing valuable insights into cellular phenotypes beyond what can be inferred from transcriptome analysis alone. However, insufficient large-scale integrated databases hinder researchers from accessing and exploring single-cell proteomics, impeding the advancement of this field. To fill this deficiency, we present a comprehensive database, namely Single-cell Proteomic DataBase (SPDB, https://scproteomicsdb.com/), for general single-cell proteomic data, including antibody-based or mass spectrometry-based single-cell proteomics. Equipped with standardized data process and a user-friendly web interface, SPDB provides unified data formats for convenient interaction with downstream analysis, and offers not only dataset-level but also protein-level data search and exploration capabilities. To enable detailed exhibition of single-cell proteomic data, SPDB also provides a module for visualizing data from the perspectives of cell metadata or protein features. The current version of SPDB encompasses 133 antibody-based single-cell proteomic datasets involving more than 300 million cells and over 800 marker/surface proteins, and 10 mass spectrometry-based single-cell proteomic datasets involving more than 4000 cells and over 7000 proteins. Overall, SPDB is envisioned to be explored as a useful resource that will facilitate the wider research communities by providing detailed insights into proteomics from the single-cell perspective.

Inhibition of Hmbox1 Promotes Cardiomyocyte Survival and Glucose Metabolism Through Gck Activation in Ischemia/Reperfusion Injury.

BACKGROUND: Exercise-induced physiological cardiac growth regulators may protect the heart from ischemia/reperfusion (I/R) injury. Homeobox-containing 1 (Hmbox1), a homeobox family member, has been identified as a putative transcriptional repressor and is downregulated in the exercised heart. However, its roles in exercise-induced physiological cardiac growth and its potential protective effects against cardiac I/R injury remain largely unexplored. METHODS: We studied the function of Hmbox1 in exercise-induced physiological cardiac growth in mice after 4 weeks of swimming exercise. Hmbox1 expression was then evaluated in human heart samples from deceased patients with myocardial infarction and in the animal cardiac I/R injury model. Its role in cardiac I/R injury was examined in mice with adeno-associated virus 9 (AAV9) vector-mediated Hmbox1 knockdown and in those with cardiac myocyte-specific Hmbox1 ablation. We performed RNA sequencing, promoter prediction, and binding assays and identified glucokinase (Gck) as a downstream effector of Hmbox1. The effects of Hmbox1 together with Gck were examined in cardiomyocytes to evaluate their cell size, proliferation, apoptosis, mitochondrial respiration, and glycolysis. The function of upstream regulator of Hmbox1, ETS1, was investigated through ETS1 overexpression in cardiac I/R mice in vivo. RESULTS: We demonstrated that Hmbox1 downregulation was required for exercise-induced physiological cardiac growth. Inhibition of Hmbox1 increased cardiomyocyte size in isolated neonatal rat cardiomyocytes and human embryonic stem cell-derived cardiomyocytes but did not affect cardiomyocyte proliferation. Under pathological conditions, Hmbox1 was upregulated in both human and animal postinfarct cardiac tissues. Furthermore, both cardiac myocyte-specific Hmbox1 knockout and AAV9-mediated Hmbox1 knockdown protected against cardiac I/R injury and heart failure. Therapeutic effects were observed when sh-Hmbox1 AAV9 was administered after I/R injury. Inhibition of Hmbox1 activated the Akt/mTOR/P70S6K pathway and transcriptionally upregulated Gck, leading to reduced apoptosis and improved mitochondrial respiration and glycolysis in cardiomyocytes. ETS1 functioned as an upstream negative regulator of Hmbox1 transcription, and its overexpression was protective against cardiac I/R injury. CONCLUSIONS: Our studies unravel a new role for the transcriptional repressor Hmbox1 in exercise-induced physiological cardiac growth. They also highlight the therapeutic potential of targeting Hmbox1 to improve myocardial survival and glucose metabolism after I/R injury.

scMHNN: a novel hypergraph neural network for integrative analysis of single-cell epigenomic, transcriptomic and proteomic data.

Technological advances have now made it possible to simultaneously profile the changes of epigenomic, transcriptomic and proteomic at the single cell level, allowing a more unified view of cellular phenotypes and heterogeneities. However, current computational tools for single-cell multi-omics data integration are mainly tailored for bi-modality data, so new tools are urgently needed to integrate tri-modality data with complex associations. To this end, we develop scMHNN to integrate single-cell multi-omics data based on hypergraph neural network. After modeling the complex data associations among various modalities, scMHNN performs message passing process on the multi-omics hypergraph, which can capture the high-order data relationships and integrate the multiple heterogeneous features. Followingly, scMHNN learns discriminative cell representation via a dual-contrastive loss in self-supervised manner. Based on the pretrained hypergraph encoder, we further introduce the pre-training and fine-tuning paradigm, which allows more accurate cell-type annotation with only a small number of labeled cells as reference. Benchmarking results on real and simulated single-cell tri-modality datasets indicate that scMHNN outperforms other competing methods on both cell clustering and cell-type annotation tasks. In addition, we also demonstrate scMHNN facilitates various downstream tasks, such as cell marker detection and enrichment analysis.

SC-AIR-BERT: a pre-trained single-cell model for predicting the antigen-binding specificity of the adaptive immune receptor.

Accurately predicting the antigen-binding specificity of adaptive immune receptors (AIRs), such as T-cell receptors (TCRs) and B-cell receptors (BCRs), is essential for discovering new immune therapies. However, the diversity of AIR chain sequences limits the accuracy of current prediction methods. This study introduces SC-AIR-BERT, a pre-trained model that learns comprehensive sequence representations of paired AIR chains to improve binding specificity prediction. SC-AIR-BERT first learns the 'language' of AIR sequences through self-supervised pre-training on a large cohort of paired AIR chains from multiple single-cell resources. The model is then fine-tuned with a multilayer perceptron head for binding specificity prediction, employing the K-mer strategy to enhance sequence representation learning. Extensive experiments demonstrate the superior AUC performance of SC-AIR-BERT compared with current methods for TCR- and BCR-binding specificity prediction.

Interpretable artificial intelligence model for accurate identification of medical conditions using immune repertoire.

Underlying medical conditions, such as cancer, kidney disease and heart failure, are associated with a higher risk for severe COVID-19. Accurate classification of COVID-19 patients with underlying medical conditions is critical for personalized treatment decision and prognosis estimation. In this study, we propose an interpretable artificial intelligence model termed VDJMiner to mine the underlying medical conditions and predict the prognosis of COVID-19 patients according to their immune repertoires. In a cohort of more than 1400 COVID-19 patients, VDJMiner accurately identifies multiple underlying medical conditions, including cancers, chronic kidney disease, autoimmune disease, diabetes, congestive heart failure, coronary artery disease, asthma and chronic obstructive pulmonary disease, with an average area under the receiver operating characteristic curve (AUC) of 0.961. Meanwhile, in this same cohort, VDJMiner achieves an AUC of 0.922 in predicting severe COVID-19. Moreover, VDJMiner achieves an accuracy of 0.857 in predicting the response of COVID-19 patients to tocilizumab treatment on the leave-one-out test. Additionally, VDJMiner interpretively mines and scores V(D)J gene segments of the T-cell receptors that are associated with the disease. The identified associations between single-cell V(D)J gene segments and COVID-19 are highly consistent with previous studies. The source code of VDJMiner is publicly accessible at https://github.com/TencentAILabHealthcare/VDJMiner. The web server of VDJMiner is available at https://gene.ai.tencent.com/VDJMiner/.

iAMPCN: a deep-learning approach for identifying antimicrobial peptides and their functional activities.

Antimicrobial peptides (AMPs) are short peptides that play crucial roles in diverse biological processes and have various functional activities against target organisms. Due to the abuse of chemical antibiotics and microbial pathogens' increasing resistance to antibiotics, AMPs have the potential to be alternatives to antibiotics. As such, the identification of AMPs has become a widely discussed topic. A variety of computational approaches have been developed to identify AMPs based on machine learning algorithms. However, most of them are not capable of predicting the functional activities of AMPs, and those predictors that can specify activities only focus on a few of them. In this study, we first surveyed 10 predictors that can identify AMPs and their functional activities in terms of the features they employed and the algorithms they utilized. Then, we constructed comprehensive AMP datasets and proposed a new deep learning-based framework, iAMPCN (identification of AMPs based on CNNs), to identify AMPs and their related 22 functional activities. Our experiments demonstrate that iAMPCN significantly improved the prediction performance of AMPs and their corresponding functional activities based on four types of sequence features. Benchmarking experiments on the independent test datasets showed that iAMPCN outperformed a number of state-of-the-art approaches for predicting AMPs and their functional activities. Furthermore, we analyzed the amino acid preferences of different AMP activities and evaluated the model on datasets of varying sequence redundancy thresholds. To facilitate the community-wide identification of AMPs and their corresponding functional types, we have made the source codes of iAMPCN publicly available at https://github.com/joy50706/iAMPCN/tree/master. We anticipate that iAMPCN can be explored as a valuable tool for identifying potential AMPs with specific functional activities for further experimental validation.

CTEC: a cross-tabulation ensemble clustering approach for single-cell RNA sequencing data analysis.

MOTIVATION: Cell-type clustering is a crucial first step for single-cell RNA-seq data analysis. However, existing clustering methods often provide different results on cluster assignments with respect to their own data pre-processing, choice of distance metrics, and strategies of feature extraction, thereby limiting their practical applications. RESULTS: We propose Cross-Tabulation Ensemble Clustering (CTEC) method that formulates two re-clustering strategies (distribution- and outlier-based) via cross-tabulation. Benchmarking experiments on five scRNA-Seq datasets illustrate that the proposed CTEC method offers significant improvements over the individual clustering methods. Moreover, CTEC-DB outperforms the state-of-the-art ensemble methods for single-cell data clustering, with 45.4% and 17.1% improvement over the single-cell aggregated from ensemble clustering method (SAFE) and the single-cell aggregated clustering via Mixture model ensemble method (SAME), respectively, on the two-method ensemble test. AVAILABILITY AND IMPLEMENTATION: The source code of the benchmark in this work is available at the GitHub repository https://github.com/LWCHN/CTEC.git.

Integrative analysis of genetics, epigenetics and RNA expression data reveal three susceptibility loci for smoking behavior in Chinese Han population.

Despite numerous studies demonstrate that genetics and epigenetics factors play important roles on smoking behavior, our understanding of their functional relevance and coordinated regulation remains largely unknown. Here we present a multiomics study on smoking behavior for Chinese smoker population with the goal of not only identifying smoking-associated functional variants but also deciphering the pathogenesis and mechanism underlying smoking behavior in this under-studied ethnic population. After whole-genome sequencing analysis of 1329 Chinese Han male samples in discovery phase and OpenArray analysis of 3744 samples in replication phase, we discovered that three novel variants located near FOXP1 (rs7635815), and between DGCR6 and PRODH (rs796774020), and in ARVCF (rs148582811) were significantly associated with smoking behavior. Subsequently cis-mQTL and cis-eQTL analysis indicated that these variants correlated significantly with the differential methylation regions (DMRs) or differential expressed genes (DEGs) located in the regions where these variants present. Finally, our in silico multiomics analysis revealed several hub genes, like DRD2, PTPRD, FOXP1, COMT, CTNNAP2, to be synergistic regulated each other in the etiology of smoking.

Harnessing computational spatial omics to explore the spatial biology intricacies.

Spatially resolved transcriptomics (SRT) has unlocked new dimensions in our understanding of intricate tissue architectures. However, this rapidly expanding field produces a wealth of diverse and voluminous data, necessitating the evolution of sophisticated computational strategies to unravel inherent patterns. Two distinct methodologies, gene spatial pattern recognition (GSPR) and tissue spatial pattern recognition (TSPR), have emerged as vital tools in this process. GSPR methodologies are designed to identify and classify genes exhibiting noteworthy spatial patterns, while TSPR strategies aim to understand intercellular interactions and recognize tissue domains with molecular and spatial coherence. In this review, we provide a comprehensive exploration of SRT, highlighting crucial data modalities and resources that are instrumental for the development of methods and biological insights. We address the complexities and challenges posed by the use of heterogeneous data in developing GSPR and TSPR methodologies and propose an optimal workflow for both. We delve into the latest advancements in GSPR and TSPR, examining their interrelationships. Lastly, we peer into the future, envisaging the potential directions and perspectives in this dynamic field.

Clarion is a multi-label problem transformation method for identifying mRNA subcellular localizations.

Subcellular localization of messenger RNAs (mRNAs) plays a key role in the spatial regulation of gene activity. The functions of mRNAs have been shown to be closely linked with their localizations. As such, understanding of the subcellular localizations of mRNAs can help elucidate gene regulatory networks. Despite several computational methods that have been developed to predict mRNA localizations within cells, there is still much room for improvement in predictive performance, especially for the multiple-location prediction. In this study, we proposed a novel multi-label multi-class predictor, termed Clarion, for mRNA subcellular localization prediction. Clarion was developed based on a manually curated benchmark dataset and leveraged the weighted series method for multi-label transformation. Extensive benchmarking tests demonstrated Clarion achieved competitive predictive performance and the weighted series method plays a crucial role in securing superior performance of Clarion. In addition, the independent test results indicate that Clarion outperformed the state-of-the-art methods and can secure accuracy of 81.47, 91.29, 79.77, 92.10, 89.15, 83.74, 80.74, 79.23 and 84.74% for chromatin, cytoplasm, cytosol, exosome, membrane, nucleolus, nucleoplasm, nucleus and ribosome, respectively. The webserver and local stand-alone tool of Clarion is freely available at http://monash.bioweb.cloud.edu.au/Clarion/.

Targeting tumor heterogeneity: multiplex-detection-based multiple instance learning for whole slide image classification.

MOTIVATION: Multiple instance learning (MIL) is a powerful technique to classify whole slide images (WSIs) for diagnostic pathology. The key challenge of MIL on WSI classification is to discover the critical instances that trigger the bag label. However, tumor heterogeneity significantly hinders the algorithm's performance. RESULTS: Here, we propose a novel multiplex-detection-based multiple instance learning (MDMIL) which targets tumor heterogeneity by multiplex detection strategy and feature constraints among samples. Specifically, the internal query generated after the probability distribution analysis and the variational query optimized throughout the training process are utilized to detect potential instances in the form of internal and external assistance, respectively. The multiplex detection strategy significantly improves the instance-mining capacity of the deep neural network. Meanwhile, a memory-based contrastive loss is proposed to reach consistency on various phenotypes in the feature space. The novel network and loss function jointly achieve high robustness towards tumor heterogeneity. We conduct experiments on three computational pathology datasets, e.g. CAMELYON16, TCGA-NSCLC, and TCGA-RCC. Benchmarking experiments on the three datasets illustrate that our proposed MDMIL approach achieves superior performance over several existing state-of-the-art methods. AVAILABILITY AND IMPLEMENTATION: MDMIL is available for academic purposes at https://github.com/ZacharyWang-007/MDMIL.

Structural and chemical study of complex silver patterns additively manufactured by multi-photon reduction.

Multi-photon reduction (MPR) based on femtosecond laser makes rapid prototyping and molding in micro-nano scale feasible, but is limited in material selectivity due to lack of the understanding of the reaction mechanism in MPR process. In this paper, additively manufacturing of complex silver-based patterns through MPR is demonstrated. The effects of laser parameters, including laser pulse energies and scanning speeds, on the structural and chemical characteristics of the printed structures are systematically investigated. The results show that the geometric size of printed cubes deviates from the designed size further by increasing laser pulse energy or decreasing scanning speed. The reaction mechanism of MPR is revealed by studying the elemental composition and chemical structures of printed cubes. The evolution of Raman spectra upon the laser processing parameters suggests that the MPR process mainly includes two processes: reduction and decomposition. In the MPR process, silver ions are reduced and grow into particles by accepting the electrons from ethonal molecules; meanwhile carboxyl groups in polyvinylpyrrolidone are decomposed and form amorphous carbon that is attached on the surface of silver particles. The conductivity of silver wires fabricated by MPR reaches 2 × 105S m-1and stays relatively constant as varying their cross section area, suggesting excellent electrical conduction. The understanding of the MPR process would accelerate the development of MPR technology and the implementation of MPR in micro-electromechanical systems could therefore be envisioned.

Study on Microstructure and High Temperature Stability of WTaVTiZrx Refractory High Entropy Alloy Prepared by Laser Cladding.

The extremely harsh environment of the high temperature plasma imposes strict requirements on the construction materials of the first wall in a fusion reactor. In this work, a refractory alloy system, WTaVTiZrx, with low activation and high entropy, was theoretically designed based on semi-empirical formula and produced using a laser cladding method. The effects of Zr proportions on the metallographic microstructure, phase composition, and alloy chemistry of a high-entropy alloy cladding layer were investigated using a metallographic microscope, XRD (X-ray diffraction), SEM (scanning electron microscope), and EDS (energy dispersive spectrometer), respectively. The high-entropy alloys have a single-phase BCC structure, and the cladding layers exhibit a typical dendritic microstructure feature. The evolution of microstructure and mechanical properties of the high-entropy alloys, with respect to annealing temperature, was studied to reveal the performance stability of the alloy at a high temperature. The microstructure of the annealed samples at 900 °C for 5-10 h did not show significant changes compared to the as-cast samples, and the microhardness increased to 988.52 HV, which was higher than that of the as-cast samples (725.08 HV). When annealed at 1100 °C for 5 h, the microstructure remained unchanged, and the microhardness increased. However, after annealing for 10 h, black substances appeared in the microstructure, and the microhardness decreased, but it was still higher than the matrix. When annealed at 1200 °C for 5-10 h, the microhardness did not increase significantly compared to the as-cast samples, and after annealing for 10 h, the microhardness was even lower than that of the as-cast samples. The phase of the high entropy alloy did not change significantly after high-temperature annealing, indicating good phase stability at high temperatures. After annealing for 10 h, the microhardness was lower than that of the as-cast samples. The phase of the high entropy alloy remained unchanged after high-temperature annealing, demonstrating good phase stability at high temperatures.

Near-Infrared II Hyperspectral Imaging Improves the Accuracy of Pathological Sampling of Multiple Cancer Types.

Pathological histology is the "gold standard" for clinical diagnosis of cancer. Incomplete or excessive sampling of the formalin-fixed excised cancer specimen will result in inaccurate histologic assessment or excessive workload. Conventionally, pathologists perform specimen sampling relying on naked-eye observation, which is subjective and limited by human perception. Precise identification of cancer tissue, size, and margin is challenging, especially for lesions with inconspicuous tumors. To overcome the limits of human eye perception (visible: 400-700 nm) and improve the sampling efficiency, in this study, we propose using a second near-infrared window (NIR-II: 900-1700 nm) hyperspectral imaging (HSI) system to assist specimen sampling on the strength of the verified deep anatomical penetration and low scattering characteristics of the NIR-II optical window. We used selected NIR-II HSI narrow bands to synthesize color images for human eye observation and also applied a machine learning-based algorithm on the complete NIR-II HSI data for automatic tissue classification to assist pathologists in specimen sampling. A total of 92 tumor samples were collected, including 7 types. Sixty-two (62/92) samples were used as the validation set. Five experienced pathologists marked the contour of the cancer tissue on conventional color images by using different methods, and compared it with the "gold standard," showing that NIR-II HSI-assisted methods had significant improvements in determining cancer tissue compared with conventional methods (conventional color image with or without X-ray). The proposed system can be easily integrated into the current workflow, with high imaging efficiency and no ionizing radiation. It may also find applications in intraoperative detection of residual lesions and identification of different tissues.

Multi-level attention graph neural network based on co-expression gene modules for disease diagnosis and prognosis.

MOTIVATION: Advanced deep learning techniques have been widely applied in disease diagnosis and prognosis with clinical omics, especially gene expression data. In the regulation of biological processes and disease progression, genes often work interactively rather than individually. Therefore, investigating gene association information and co-functional gene modules can facilitate disease state prediction. RESULTS: To explore the gene modules and inter-gene relational information contained in the omics data, we propose a novel multi-level attention graph neural network (MLA-GNN) for disease diagnosis and prognosis. Specifically, we format omics data into co-expression graphs via weighted correlation network analysis, and then construct multi-level graph features, finally fuse them through a well-designed multi-level graph feature fully fusion module to conduct predictions. For model interpretation, a novel full-gradient graph saliency mechanism is developed to identify the disease-relevant genes. MLA-GNN achieves state-of-the-art performance on transcriptomic data from TCGA-LGG/TCGA-GBM and proteomic data from coronavirus disease 2019 (COVID-19)/non-COVID-19 patient sera. More importantly, the relevant genes selected by our model are interpretable and are consistent with the clinical understanding. AVAILABILITYAND IMPLEMENTATION: The codes are available at https://github.com/TencentAILabHealthcare/MLA-GNN.

A machine learning approach for predicting perihematomal edema expansion in patients with intracerebral hemorrhage.

OBJECTIVES: Preventing the expansion of perihematomal edema (PHE) represents a novel strategy for the improvement of neurological outcomes in intracerebral hemorrhage (ICH) patients. Our goal was to predict early and delayed PHE expansion using a machine learning approach. METHODS: We enrolled 550 patients with spontaneous ICH to study early PHE expansion, and 389 patients to study delayed expansion. Two imaging researchers rated the shape and density of hematoma in non-contrast computed tomography (NCCT). We trained a radiological machine learning (ML) model, a radiomics ML model, and a combined ML model, using data from radiomics, traditional imaging, and clinical indicators. We then validated these models on an independent dataset by using a nested 4-fold cross-validation approach. We compared models with respect to their predictive performance, which was assessed using the receiver operating characteristic curve. RESULTS: For both early and delayed PHE expansion, the combined ML model was most predictive (early/delayed AUC values were 0.840/0.705), followed by the radiomics ML model (0.799/0.663), the radiological ML model (0.779/0.631), and the imaging readers (reader 1: 0.668/0.565, reader 2: 0.700/0.617). CONCLUSION: We validated a machine learning approach with high interpretability for the prediction of early and delayed PHE expansion. This new technique may assist clinical practice for the management of neurocritical patients with ICH. KEY POINTS: • This is the first study to use artificial intelligence technology for the prediction of perihematomal edema expansion. • A combined machine learning model, trained on data from radiomics, clinical indicators, and imaging features associated with hematoma expansion, outperformed all other methods.