Systematic analysis on the horse-shoe-like effect in PCA plots of scRNA-seq data.

Motivation: In single-cell studies, principal component analysis (PCA) is widely used to reduce the dimensionality of dataset and visualize in 2D or 3D PC plots. Scientists often focus on different clusters within PC plot, overlooking the specific phenomenon, such as horse-shoe-like effect, that may reveal hidden knowledge about underlying biological dataset. This phenomenon remains largely unexplored in single-cell studies. Results: In this study, we investigated into the horse-shoe-like effect in PC plots using simulated and real scRNA-seq datasets. We systematically explain horse-shoe-like phenomenon from various inter-related perspectives. Initially, we establish an intuitive understanding with the help of simulated datasets. Then, we generalized the acquired knowledge on real biological scRNA-seq data. Experimental results provide logical explanations and understanding for the appearance of horse-shoe-like effect in PC plots. Furthermore, we identify a potential problem with a well-known theory of 'distance saturation property' attributed to induce horse-shoe phenomenon. Finally, we analyse a mathematical model for horse-shoe effect that suggests trigonometric solutions to estimated eigenvectors. We observe significant resemblance after comparing the results of mathematical model with simulated and real scRNA-seq datasets. Availability and implementation: The code for reproducing the results of this study is available at: https://github.com/najeebullahshah/PCA-Horse-Shoe.

Dynamic regulation of innate lymphoid cell development during ontogeny.

The helper-like ILC contains various functional subsets, such as ILC1, ILC2, ILC3 and LTi cells, mediating the immune responses against viruses, parasites, and extracellular bacteria, respectively. Among them, LTi cells are also crucial for the formation of peripheral lymphoid tissues, such as lymph nodes. Our research, along with others', indicates a high proportion of LTi cells in the fetal ILC pool, which significantly decreases after birth. Conversely, the proportion of non-LTi ILCs increases postnatally, corresponding to the need for LTi cells to mediate lymphoid tissue formation during fetal stages and other ILC subsets to combat diverse pathogen infections postnatally. However, the regulatory mechanism for this transition remains unclear. In this study, we observed a preference for fetal ILC progenitors to differentiate into LTi cells, while postnatal bone marrow ILC progenitors preferentially differentiate into non-LTi ILCs. Particularly, this differentiation shift occurs within the first week after birth in mice. Further analysis revealed that adult ILC progenitors exhibit stronger activation of the Notch signaling pathway compared to fetal counterparts, accompanied by elevated Gata3 expression and decreased Rorc expression, leading to a transition from fetal LTi cell-dominant states to adult non-LTi ILC-dominant states. This study suggests that the body can regulate ILC development by modulating the activation level of the Notch signaling pathway, thereby acquiring different ILC subsets to accommodate the varying demands within the body at different developmental stages.

Building a learnable universal coordinate system for single-cell atlas with a joint-VAE model.

A universal coordinate system that can ensemble the huge number of cells and capture their heterogeneities is of vital importance for constructing large-scale cell atlases as references for molecular and cellular studies. Studies have shown that cells exhibit multifaceted heterogeneities in their transcriptomic features at multiple resolutions. This nature of complexity makes it hard to design a fixed coordinate system through a combination of known features. It is desirable to build a learnable universal coordinate model that can capture major heterogeneities and serve as a controlled generative model for data augmentation. We developed UniCoord, a specially-tuned joint-VAE model to represent single-cell transcriptomic data in a lower-dimensional latent space with high interpretability. Each latent dimension can represent either discrete or continuous feature, and either supervised by prior knowledge or unsupervised. The latent dimensions can be easily reconfigured to generate pseudo transcriptomic profiles with desired properties. UniCoord can also be used as a pre-trained model to analyze new data with unseen cell types and thus can serve as a feasible framework for cell annotation and comparison. UniCoord provides a prototype for a learnable universal coordinate framework to enable better analysis and generation of cells with highly orchestrated functions and heterogeneities.

A brain cell atlas integrating single-cell transcriptomes across human brain regions.

While single-cell technologies have greatly advanced our comprehension of human brain cell types and functions, studies including large numbers of donors and multiple brain regions are needed to extend our understanding of brain cell heterogeneity. Integrating atlas-level single-cell data presents a chance to reveal rare cell types and cellular heterogeneity across brain regions. Here we present the Brain Cell Atlas, a comprehensive reference atlas of brain cells, by assembling single-cell data from 70 human and 103 mouse studies of the brain throughout major developmental stages across brain regions, covering over 26.3 million cells or nuclei from both healthy and diseased tissues. Using machine-learning based algorithms, the Brain Cell Atlas provides a consensus cell type annotation, and it showcases the identification of putative neural progenitor cells and a cell subpopulation of PCDH9high microglia in the human brain. We demonstrate the gene regulatory difference of PCDH9high microglia between hippocampus and prefrontal cortex and elucidate the cell-cell communication network. The Brain Cell Atlas presents an atlas-level integrative resource for comparing brain cells in different environments and conditions within the Human Cell Atlas.

scDiffusion: conditional generation of high-quality single-cell data using diffusion model.

MOTIVATION: Single-cell RNA sequencing (scRNA-seq) data are important for studying the laws of life at single-cell level. However, it is still challenging to obtain enough high-quality scRNA-seq data. To mitigate the limited availability of data, generative models have been proposed to computationally generate synthetic scRNA-seq data. Nevertheless, the data generated with current models are not very realistic yet, especially when we need to generate data with controlled conditions. In the meantime, diffusion models have shown their power in generating data with high fidelity, providing a new opportunity for scRNA-seq generation. RESULTS: In this study, we developed scDiffusion, a generative model combining the diffusion model and foundation model to generate high-quality scRNA-seq data with controlled conditions. We designed multiple classifiers to guide the diffusion process simultaneously, enabling scDiffusion to generate data under multiple condition combinations. We also proposed a new control strategy called Gradient Interpolation. This strategy allows the model to generate continuous trajectories of cell development from a given cell state. Experiments showed that scDiffusion could generate single-cell gene expression data closely resembling real scRNA-seq data. Also, scDiffusion can conditionally produce data on specific cell types including rare cell types. Furthermore, we could use the multiple-condition generation of scDiffusion to generate cell type that was out of the training data. Leveraging the Gradient Interpolation strategy, we generated a continuous developmental trajectory of mouse embryonic cells. These experiments demonstrate that scDiffusion is a powerful tool for augmenting the real scRNA-seq data and can provide insights into cell fate research. AVAILABILITY AND IMPLEMENTATION: scDiffusion is openly available at the GitHub repository https://github.com/EperLuo/scDiffusion or Zenodo https://zenodo.org/doi/10.5281/zenodo.13268742.

HCCDB v2.0: Decompose Expression Variations by Single-cell RNA-seq and Spatial Transcriptomics in HCC.

Large-scale transcriptomic data are crucial for understanding the molecular features of hepatocellular carcinoma (HCC). Integrated 15 transcriptomic datasets of HCC clinical samples, the first version of HCC database (HCCDB v1.0) was released in 2018. Through the meta-analysis of differentially expressed genes and prognosis-related genes across multiple datasets, it provides a systematic view of the altered biological processes and the inter-patient heterogeneities of HCC with high reproducibility and robustness. With four years having passed, the database now needs integration of recently published datasets. Furthermore, the latest single-cell and spatial transcriptomics have provided a great opportunity to decipher complex gene expression variations at the cellular level with spatial architecture. Here, we present HCCDB v2.0, an updated version that combines bulk, single-cell, and spatial transcriptomic data of HCC clinical samples. It dramatically expands the bulk sample size by adding 1656 new samples from 11 datasets to the existing 3917 samples, thereby enhancing the reliability of transcriptomic meta-analysis. A total of 182,832 cells and 69,352 spatial spots are added to the single-cell and spatial transcriptomics sections, respectively. A novel single-cell level and 2-dimension (sc-2D) metric is proposed as well to summarize cell type-specific and dysregulated gene expression patterns. Results are all graphically visualized in our online portal, allowing users to easily retrieve data through a user-friendly interface and navigate between different views. With extensive clinical phenotypes and transcriptomic data in the database, we show two applications for identifying prognosis-associated cells and tumor microenvironment. HCCDB v2.0 is available at http://lifeome.net/database/hccdb2.

Reciprocal regulation of T follicular helper cells and dendritic cells drives colitis development.

The immunological mechanisms underlying chronic colitis are poorly understood. T follicular helper (TFH) cells are critical in helping B cells during germinal center reactions. In a T cell transfer colitis model, a lymphoid structure composed of mature dendritic cells (DCs) and TFH cells was found within T cell zones of colonic lymphoid follicles. TFH cells were required for mature DC accumulation, the formation of DC-T cell clusters and colitis development. Moreover, DCs promoted TFH cell differentiation, contributing to colitis development. A lineage-tracing analysis showed that, following migration to the lamina propria, TFH cells transdifferentiated into long-lived pathogenic TH1 cells, promoting colitis development. Our findings have therefore demonstrated the reciprocal regulation of TFH cells and DCs in colonic lymphoid follicles, which is critical in chronic colitis pathogenesis.

Large-scale foundation model on single-cell transcriptomics.

Large pretrained models have become foundation models leading to breakthroughs in natural language processing and related fields. Developing foundation models for deciphering the 'languages' of cells and facilitating biomedical research is promising yet challenging. Here we developed a large pretrained model scFoundation, also named 'xTrimoscFoundationα', with 100 million parameters covering about 20,000 genes, pretrained on over 50 million human single-cell transcriptomic profiles. scFoundation is a large-scale model in terms of the size of trainable parameters, dimensionality of genes and volume of training data. Its asymmetric transformer-like architecture and pretraining task design empower effectively capturing complex context relations among genes in a variety of cell types and states. Experiments showed its merit as a foundation model that achieved state-of-the-art performances in a diverse array of single-cell analysis tasks such as gene expression enhancement, tissue drug response prediction, single-cell drug response classification, single-cell perturbation prediction, cell type annotation and gene module inference.

Benchmarking multi-omics integration algorithms across single-cell RNA and ATAC data.

Recent advancements in single-cell sequencing technologies have generated extensive omics data in various modalities and revolutionized cell research, especially in the single-cell RNA and ATAC data. The joint analysis across scRNA-seq data and scATAC-seq data has paved the way to comprehending the cellular heterogeneity and complex cellular regulatory networks. Multi-omics integration is gaining attention as an important step in joint analysis, and the number of computational tools in this field is growing rapidly. In this paper, we benchmarked 12 multi-omics integration methods on three integration tasks via qualitative visualization and quantitative metrics, considering six main aspects that matter in multi-omics data analysis. Overall, we found that different methods have their own advantages on different aspects, while some methods outperformed other methods in most aspects. We therefore provided guidelines for selecting appropriate methods for specific scenarios and tasks to help obtain meaningful insights from multi-omics data integration.

Label Informed Contrastive Pretraining for Node Importance Estimation on Knowledge Graphs.

Node importance estimation (NIE) is the task of inferring the importance scores of the nodes in a graph. Due to the availability of richer data and knowledge, recent research interests of NIE have been dedicated to knowledge graphs (KGs) for predicting future or missing node importance scores. Existing state-of-the-art NIE methods train the model by available labels, and they consider every interested node equally before training. However, the nodes with higher importance often require or receive more attention in real-world scenarios, e.g., people may care more about the movies or webpages with higher importance. To this end, we introduce Label Informed ContrAstive Pretraining (LICAP) to the NIE problem for being better aware of the nodes with high importance scores. Specifically, LICAP is a novel type of contrastive learning (CL) framework that aims to fully utilize continuous labels to generate contrastive samples for pretraining embeddings. Considering the NIE problem, LICAP adopts a novel sampling strategy called top nodes preferred hierarchical sampling to first group all interested nodes into a top bin and a nontop bin based on node importance scores, and then divide the nodes within the top bin into several finer bins also based on the scores. The contrastive samples are generated from those bins and are then used to pretrain node embeddings of KGs via a newly proposed predicate-aware graph attention networks (PreGATs), so as to better separate the top nodes from nontop nodes, and distinguish the top nodes within the top bin by keeping the relative order among finer bins. Extensive experiments demonstrate that the LICAP pretrained embeddings can further boost the performance of existing NIE methods and achieve new state-of-the-art performance regarding both regression and ranking metrics. The source code for reproducibility is available at https://github.com/zhangtia16/LICAP.

scDecouple: decoupling cellular response from infected proportion bias in scCRISPR-seq.

Single-cell clustered regularly interspaced short palindromic repeats-sequencing (scCRISPR-seq) is an emerging high-throughput CRISPR screening technology where the true cellular response to perturbation is coupled with infected proportion bias of guide RNAs (gRNAs) across different cell clusters. The mixing of these effects introduces noise into scCRISPR-seq data analysis and thus obstacles to relevant studies. We developed scDecouple to decouple true cellular response of perturbation from the influence of infected proportion bias. scDecouple first models the distribution of gene expression profiles in perturbed cells and then iteratively finds the maximum likelihood of cell cluster proportions as well as the cellular response for each gRNA. We demonstrated its performance in a series of simulation experiments. By applying scDecouple to real scCRISPR-seq data, we found that scDecouple enhances the identification of biologically perturbation-related genes. scDecouple can benefit scCRISPR-seq data analysis, especially in the case of heterogeneous samples or complex gRNA libraries.

STEM enables mapping of single-cell and spatial transcriptomics data with transfer learning.

Profiling spatial variations of cellular composition and transcriptomic characteristics is important for understanding the physiology and pathology of tissues. Spatial transcriptomics (ST) data depict spatial gene expression but the currently dominating high-throughput technology is yet not at single-cell resolution. Single-cell RNA-sequencing (SC) data provide high-throughput transcriptomic information at the single-cell level but lack spatial information. Integrating these two types of data would be ideal for revealing transcriptomic landscapes at single-cell resolution. We develop the method STEM (SpaTially aware EMbedding) for this purpose. It uses deep transfer learning to encode both ST and SC data into a unified spatially aware embedding space, and then uses the embeddings to infer SC-ST mapping and predict pseudo-spatial adjacency between cells in SC data. Semi-simulation and real data experiments verify that the embeddings preserved spatial information and eliminated technical biases between SC and ST data. We apply STEM to human squamous cell carcinoma and hepatic lobule datasets to uncover the localization of rare cell types and reveal cell-type-specific gene expression variation along a spatial axis. STEM is powerful for mapping SC and ST data to build single-cell level spatial transcriptomic landscapes, and can provide mechanistic insights into the spatial heterogeneity and microenvironments of tissues.

Single-cell transcriptomic analysis reveals tumor cell heterogeneity and immune microenvironment features of pituitary neuroendocrine tumors.

BACKGROUND: Pituitary neuroendocrine tumors (PitNETs) are one of the most common types of intracranial tumors. Currently, the cellular characteristics of normal pituitary and various other types of PitNETs are still not completely understood. METHODS: We performed single-cell RNA sequencing (scRNA-seq) on 4 normal samples and 24 PitNET samples for comprehensive bioinformatics analysis. Findings regarding the function of PBK in the aggressive tumor cells were validated by siRNA knockdown, overexpression, and transwell experiments. RESULTS: We first constructed a reference cell atlas of the human pituitary. Subsequent scRNA-seq analysis of PitNET samples, representing major tumor subtypes, shed light on the intrinsic cellular heterogeneities of the tumor cells and tumor microenvironment (TME). We found that the expression of hormone-encoding genes defined the major variations of the PIT1-lineage tumor cell transcriptomic heterogeneities. A sub-population of TPIT-lineage tumor cells highly expressing GZMK suggested a novel subtype of corticotroph tumors. In immune cells, we found two clusters of tumor-associated macrophages, which were both highly enriched in PitNETs but with distinct functional characteristics. In PitNETs, the stress response pathway was significantly activated in T cells. While a majority of these tumors are benign, our study unveils a common existence of aggressive tumor cells in the studied samples, which highly express a set of malignant signature genes. The following functional experiments confirmed the oncogenic role of selected up-regulated genes. The over-expression of PBK could promote both tumor cell proliferation and migration, and it was also significantly associated with poor prognosis in PitNET patients. CONCLUSIONS: Our data and analysis manifested the basic cell types in the normal pituitary and inherent heterogeneity of PitNETs, identified several features of the tumor immune microenvironments, and found a novel epithelial cell sub-population with aggressive signatures across all the studied cases.

The spatial and single-cell analysis reveals remodeled immune microenvironment induced by synthetic oncolytic adenovirus treatment.

Oncolytic viruses are multifaceted tumor killers, which can function as tumor vaccines to boost systemic antitumor immunity. In previous study, we rationally designed a synthetic oncolytic adenovirus (SynOV) harboring a synthetic gene circuit, which can kill tumors in mouse hepatocellular carcinoma (HCC) models. In this study, we demonstrated that SynOV could sense the tumor biomarkers to lyse tumors in a dosage-dependent manner, and killed PD-L1 antibody resistant tumor cells in mouse model. Meanwhile, we observed SynOV could cure liver cancer and partially alleviate the liver cancer with distant metastasis by activating systemic antitumor immunity. To understand its high efficacy, it is essential to explore the cellular and molecular features of the remodeled tumor microenvironment (TME). By combining spatial transcriptome sequencing and single-cell RNA sequencing, we successfully depicted the remodeled TME at single cell resolution. The state transition of immune cells and stromal cells towards an antitumor and normalized status exemplified the overall cancer-suppressive TME after SynOV treatment. Specifically, SynOV treatment increased the proportion of CD8+ T cells, enhanced the cell-cell communication of Cxcl9-Cxcr3, and normalized the Kupffer cells and macrophages in the TME. Furthermore, we observed that SynOV could induce distant responses to reduce tumor burden in metastatic HCC patient in the Phase I clinical trial. In summary, our results suggest that SynOV can trigger systemic antitumor immunity to induce CD8+ T cells and normalize the abundance of immune cells to remodel the TME, which promises a powerful option to treat HCC in the future.

Lack of incremental value of three-dimensional measurement in assessing invasiveness for lung cancer.

OBJECTIVES: The aim of this study was to evaluate the performance of consolidation-to-tumour ratio (CTR) and the radiomic models in two- and three-dimensional modalities for assessing radiological invasiveness in early-stage lung adenocarcinoma. METHODS: A retrospective analysis was conducted on patients with early-stage lung adenocarcinoma from Guangdong Provincial People's Hospital and Shenzhen People's Hospital. Manual delineation of pulmonary nodules along the boundary was performed on cross-sectional images to extract radiomic features. Clinicopathological characteristics and radiomic signatures were identified in both cohorts. CTR and radiomic score for every patient were calculated. The performance of CTR and radiomic models were tested and validated in the respective cohorts. RESULTS: A total of 818 patients from Guangdong Provincial People's Hospital were included in the primary cohort, while 474 patients from Shenzhen People's Hospital constituted an independent validation cohort. Both CTR and radiomic score were identified as independent factors for predicting pathological invasiveness. CTR in two- and three-dimensional modalities exhibited comparable results with areas under the receiver operating characteristic curves and were demonstrated in the validation cohort (area under the curve: 0.807 vs 0.826, P = 0.059) Furthermore, both CTR in two- and three-dimensional modalities was able to stratify patients with significant relapse-free survival (P < 0.000 vs P < 0.000) and overall survival (P = 0.003 vs P = 0.001). The radiomic models in two- and three-dimensional modalities demonstrated favourable discrimination and calibration in independent cohorts (P = 0.189). CONCLUSIONS: Three-dimensional measurement provides no additional clinical benefit compared to two-dimensional.

Latent feature extraction with a prior-based self-attention framework for spatial transcriptomics.

Rapid advances in spatial transcriptomics (ST) have revolutionized the interrogation of spatial heterogeneity and increase the demand for comprehensive methods to effectively characterize spatial domains. As a prerequisite for ST data analysis, spatial domain characterization is a crucial step for downstream analyses and biological implications. Here we propose a prior-based self-attention framework for spatial transcriptomics (PAST), a variational graph convolutional autoencoder for ST, which effectively integrates prior information via a Bayesian neural network, captures spatial patterns via a self-attention mechanism, and enables scalable application via a ripple walk sampler strategy. Through comprehensive experiments on data sets generated by different technologies, we show that PAST can effectively characterize spatial domains and facilitate various downstream analyses, including ST visualization, spatial trajectory inference and pseudotime analysis. Also, we highlight the advantages of PAST for multislice joint embedding and automatic annotation of spatial domains in newly sequenced ST data. Compared with existing methods, PAST is the first ST method that integrates reference data to analyze ST data. We anticipate that PAST will open up new avenues for researchers to decipher ST data with customized reference data, which expands the applicability of ST technology.

Decoding functional cell-cell communication events by multi-view graph learning on spatial transcriptomics.

Cell-cell communication events (CEs) are mediated by multiple ligand-receptor (LR) pairs. Usually only a particular subset of CEs directly works for a specific downstream response in a particular microenvironment. We name them as functional communication events (FCEs) of the target responses. Decoding FCE-target gene relations is: important for understanding the mechanisms of many biological processes, but has been intractable due to the mixing of multiple factors and the lack of direct observations. We developed a method HoloNet for decoding FCEs using spatial transcriptomic data by integrating LR pairs, cell-type spatial distribution and downstream gene expression into a deep learning model. We modeled CEs as a multi-view network, developed an attention-based graph learning method to train the model for generating target gene expression with the CE networks, and decoded the FCEs for specific downstream genes by interpreting trained models. We applied HoloNet on three Visium datasets of breast cancer and liver cancer. The results detangled the multiple factors of FCEs by revealing how LR signals and cell types affect specific biological processes, and specified FCE-induced effects in each single cell. We conducted simulation experiments and showed that HoloNet is more reliable on LR prioritization in comparison with existing methods. HoloNet is a powerful tool to illustrate cell-cell communication landscapes and reveal vital FCEs that shape cellular phenotypes. HoloNet is available as a Python package at https://github.com/lhc17/HoloNet.

ESICCC as a systematic computational framework for evaluation, selection, and integration of cell-cell communication inference methods.

Cell-cell communication (CCC) is critical for determining cell fates and functions in multicellular organisms. With the advent of single-cell RNA-sequencing (scRNA-seq) and spatial transcriptomics (ST), an increasing number of CCC inference methods have been developed. Nevertheless, a thorough comparison of their performances is yet to be conducted. To fill this gap, we developed a systematic benchmark framework called ESICCC to evaluate 18 ligand-receptor (LR) inference methods and five ligand/receptor-target inference methods using a total of 116 data sets, including 15 ST data sets, 15 sets of cell line perturbation data, two sets of cell type-specific expression/proteomics data, and 84 sets of sampled or unsampled scRNA-seq data. We evaluated and compared the agreement, accuracy, robustness, and usability of these methods. Regarding accuracy evaluation, RNAMagnet, CellChat, and scSeqComm emerge as the three best-performing methods for intercellular ligand-receptor inference based on scRNA-seq data, whereas stMLnet and HoloNet are the best methods for predicting ligand/receptor-target regulation using ST data. To facilitate the practical applications, we provide a decision-tree-style guideline for users to easily choose best tools for their specific research concerns in CCC inference, and develop an ensemble pipeline CCCbank that enables versatile combinations of methods and databases. Moreover, our comparative results also uncover several critical influential factors for CCC inference, such as prior interaction information, ligand-receptor scoring algorithm, intracellular signaling complexity, and spatial relationship, which may be considered in the future studies to advance the development of new methodologies.

Detection and subtyping of hepatic echinococcosis from plain CT images with deep learning: a retrospective, multicentre study.

BACKGROUND: Hepatic echinococcosis is a severe endemic disease in some underdeveloped rural areas worldwide. Qualified physicians are in short supply in such areas, resulting in low rates of accurate diagnosis of this condition. In this study, we aimed to develop and evaluate an artificial intelligence (AI) system for automated detection and subtyping of hepatic echinococcosis using plain CT images with the goal of providing interpretable assistance to radiologists and clinicians. METHODS: We developed EDAM, an echinococcosis diagnostic AI system, to provide accurate and generalisable CT analysis for distinguishing hepatic echinococcosis from hepatic cysts and normal controls (no liver lesions), as well as subtyping hepatic echinococcosis as alveolar or cystic echinococcosis. EDAM includes a slice-level prediction model for lesion classification and segmentation and a patient-level diagnostic model for patient classification. We collected a plain CT database (n=700: 395 cystic echinococcosis, 122 alveolar echinococcosis, 130 hepatic cysts, and 53 normal controls) for developing EDAM, and two additional independent cohorts (n=156) for external validation of its performance and generalisation ability. We compared the performance of EDAM with 52 experienced radiologists in diagnosing and subtyping hepatic echinococcosis. FINDINGS: EDAM showed reliable performance in patient-level diagnosis on both the internal testing data (overall area under the receiver operating characteristic curve [AUC]: 0·974 [95% CI 0·936-0·994]; accuracy: 0·952 [0·939-0·965] for cystic echinococcosis, 0·981 [0·973-0·989] for alveolar echinococcosis; sensitivity: 0·966 [0·951-0·979] for cystic echinococcosis, 0·944 [0·908-0·970] for alveolar echinococcosis) and the external testing set (overall AUC: 0·953 [95% CI 0·840-0·973]; accuracy: 0·929 [0·915-0·947] for cystic echinococcosis, 0·936 [0·919-0·950] for alveolar echinococcosis; sensitivity: 0·913 [0·879-0·944] for cystic echinococcosis, 0·868 [0·841-0·897] for alveolar echinococcosis). The sensitivity of EDAM was robust across images from different CT manufacturers. EDAM outperformed most of the enrolled radiologists in detecting both alveolar echinococcosis and cystic echinococcosis. INTERPRETATION: EDAM is a clinically applicable AI system that can provide patient-level diagnoses with interpretable results. The accuracy and generalisation ability of EDAM demonstrates its potential for clinical use, especially in underdeveloped areas. FUNDING: Project of Qinghai Provincial Department of Science and Technology of China, National Natural Science Foundation of China, and Tsinghua-Fuzhou Institute of Data Technology Project. TRANSLATION: For the Chinese translation of the abstract see Supplementary Materials section.