A novel multiscale framework for delineating cancer evolution from subclonal compositions.

OBJECTIVE: Owing to the heterogeneity in the evolution of cancer, distinguishing between diverse growth patterns and predicting long-term outcomes based on short-term measurements poses a great challenge. METHODS: A novel multiscale framework is proposed to unravel the connections between the population dynamics of cancer growth (i.e., aggressive, bounded, and indolent) and the cellular-subclonal dynamics of cancer evolution. This framework employs the non-negative lasso (NN-LASSO) algorithm to forge a link between an ordinary differential equation (ODE)-based population model and a cellular evolution model. RESULTS: The findings of our current work not only affirm the impact of subclonal composition on growth dynamics but also identify two significant subclones within heterogeneous growth patterns. Moreover, the subclonal compositions at the initial time are able to accurately discriminate diverse growth patterns through a machine learning algorithm. CONCLUSION: The proposed multiscale framework successfully delineates the intricate landscape of cancer evolution, bridging the gap between long-term growth dynamics and short-term measurements, both in simulated and real-world data. This methodology provides a novel avenue for thorough exploration into the realm of cancer evolution.

scAB detects multiresolution cell states with clinical significance by integrating single-cell genomics and bulk sequencing data.

Although single-cell sequencing has provided a powerful tool to deconvolute cellular heterogeneity of diseases like cancer, extrapolating clinical significance or identifying clinically-relevant cells remains challenging. Here, we propose a novel computational method scAB, which integrates single-cell genomics data with clinically annotated bulk sequencing data via a knowledge- and graph-guided matrix factorization model. Once combined, scAB provides a coarse- and fine-grain multiresolution perspective of phenotype-associated cell states and prognostic signatures previously not visible by single-cell genomics. We use scAB to enhance live cancer single-cell RNA-seq data, identifying clinically-relevant previously unrecognized cancer and stromal cell subsets whose signatures show a stronger poor-survival association. The identified fine-grain cell subsets are associated with distinct cancer hallmarks and prognosis power. Furthermore, scAB demonstrates its utility as a biomarker identification tool, with the ability to predict immunotherapy, drug responses and survival when applied to melanoma single-cell RNA-seq datasets and glioma single-cell ATAC-seq datasets. Across multiple single-cell and bulk datasets from different cancer types, we also demonstrate the superior performance of scAB in generating prognosis signatures and survival predictions over existing models. Overall, scAB provides an efficient tool for prioritizing clinically-relevant cell subsets and predictive signatures, utilizing large publicly available databases to improve prognosis and treatments.

Network Evolution Model-based prediction of tumor mutation burden from radiomic-clinical features in endometrial cancers.

BACKGROUND: Endometrial Cancer (EC) is one of the most prevalent malignancies that affect the female population globally. In the context of immunotherapy, Tumor Mutation Burden (TMB) in the DNA polymerase epsilon (POLE) subtype of this cancer holds promise as a viable therapeutic target. METHODS: We devised a method known as NEM-TIE to forecast the TMB status of patients with endometrial cancer. This approach utilized a combination of the Network Evolution Model, Transfer Information Entropy, Clique Percolation (CP) methodology, and Support Vector Machine (SVM) classification. To construct the Network Evolution Model, we employed an adjacency matrix that utilized transfer information entropy to assess the information gain between nodes of radiomic-clinical features. Subsequently, using the CP algorithm, we unearthed potentially pivotal modules in the Network Evolution Model. Finally, the SVM classifier extracted essential features from the module set. RESULTS: Upon analyzing the importance of modules, we discovered that the dependence count energy in tumor volumes-of-interest holds immense significance in distinguishing TMB statuses among patients with endometrial cancer. Using the 13 radiomic-clinical features extracted via NEM-TIE, we demonstrated that the area under the receiver operating characteristic curve (AUROC) in the test set is 0.98 (95% confidence interval: 0.95-1.00), surpassing the performance of existing techniques such as the mRMR and Laplacian methods. CONCLUSIONS: Our study proposed the NEM-TIE method as a means to identify the TMB status of patients with endometrial cancer. The integration of radiomic-clinical data utilizing the NEM-TIE method may offer a novel technology for supplementary diagnosis.

Oligodendrocytes that survive acute coronavirus infection induce prolonged inflammatory responses in the CNS.

Neurotropic strains of mouse hepatitis virus (MHV), a coronavirus, cause acute and chronic demyelinating encephalomyelitis with similarities to the human disease multiple sclerosis. Here, using a lineage-tracking system, we show that some cells, primarily oligodendrocytes (OLs) and oligodendrocyte precursor cells (OPCs), survive the acute MHV infection, are associated with regions of demyelination, and persist in the central nervous system (CNS) for at least 150 d. These surviving OLs express major histocompatibility complex (MHC) class I and other genes associated with an inflammatory response. Notably, the extent of inflammatory cell infiltration was variable, dependent on anatomic location within the CNS, and without obvious correlation with numbers of surviving cells. We detected more demyelination in regions with larger numbers of T cells and microglia/macrophages compared to those with fewer infiltrating cells. Conversely, in regions with less inflammation, these previously infected OLs more rapidly extended processes, consistent with normal myelinating function. Together, these results show that OLs are inducers as well as targets of the host immune response and demonstrate how a CNS infection, even after resolution, can induce prolonged inflammatory changes with CNS region-dependent impairment in remyelination.

Microglia depletion exacerbates demyelination and impairs remyelination in a neurotropic coronavirus infection.

Microglia are considered both pathogenic and protective during recovery from demyelination, but their precise role remains ill defined. Here, using an inhibitor of colony stimulating factor 1 receptor (CSF1R), PLX5622, and mice infected with a neurotropic coronavirus (mouse hepatitis virus [MHV], strain JHMV), we show that depletion of microglia during the time of JHMV clearance resulted in impaired myelin repair and prolonged clinical disease without affecting the kinetics of virus clearance. Microglia were required only during the early stages of remyelination. Notably, large deposits of extracellular vesiculated myelin and cellular debris were detected in the spinal cords of PLX5622-treated and not control mice, which correlated with decreased numbers of oligodendrocytes in demyelinating lesions in drug-treated mice. Furthermore, gene expression analyses demonstrated differential expression of genes involved in myelin debris clearance, lipid and cholesterol recycling, and promotion of oligodendrocyte function. The results also demonstrate that microglial functions affected by depletion could not be compensated by infiltrating macrophages. Together, these results demonstrate that microglia play key roles in debris clearance and in the initiation of remyelination following infection with a neurotropic coronavirus but are not necessary during later stages of remyelination.

Data-driven multi-scale mathematical modeling of SARS-CoV-2 infection reveals heterogeneity among COVID-19 patients.

Patients with coronavirus disease 2019 (COVID-19) often exhibit diverse disease progressions associated with various infectious ability, symptoms, and clinical treatments. To systematically and thoroughly understand the heterogeneous progression of COVID-19, we developed a multi-scale computational model to quantitatively understand the heterogeneous progression of COVID-19 patients infected with severe acute respiratory syndrome (SARS)-like coronavirus (SARS-CoV-2). The model consists of intracellular viral dynamics, multicellular infection process, and immune responses, and was formulated using a combination of differential equations and stochastic modeling. By integrating multi-source clinical data with model analysis, we quantified individual heterogeneity using two indexes, i.e., the ratio of infected cells and incubation period. Specifically, our simulations revealed that increasing the host antiviral state or virus induced type I interferon (IFN) production rate can prolong the incubation period and postpone the transition from asymptomatic to symptomatic outcomes. We further identified the threshold dynamics of T cell exhaustion in the transition between mild-moderate and severe symptoms, and that patients with severe symptoms exhibited a lack of naïve T cells at a late stage. In addition, we quantified the efficacy of treating COVID-19 patients and investigated the effects of various therapeutic strategies. Simulations results suggested that single antiviral therapy is sufficient for moderate patients, while combination therapies and prevention of T cell exhaustion are needed for severe patients. These results highlight the critical roles of IFN and T cell responses in regulating the stage transition during COVID-19 progression. Our study reveals a quantitative relationship underpinning the heterogeneity of transition stage during COVID-19 progression and can provide a potential guidance for personalized therapy in COVID-19 patients.

A network-based computational framework to predict and differentiate functions for gene isoforms using exon-level expression data.

MOTIVATION: Alternative splicing makes significant contributions to functional diversity of transcripts and proteins. Many alternatively spliced gene isoforms have been shown to perform specific biological functions under different contexts. In addition to gene-level expression, the advances of high-throughput sequencing offer a chance to estimate isoform-specific exon expression with a high resolution, which is informative for studying splice variants with network analysis. RESULTS: In this study, we propose a novel network-based analysis framework to predict isoform-specific functions from exon-level RNA-Seq data. In particular, based on exon-level expression data, we firstly propose a unified framework, referred to as Iso-Net, to integrate two new mathematical methods (named MINet and RVNet) that infer co-expression networks at different data scenarios. We demonstrate the superior prediction accuracy of Iso-Net over the existing methods for most simulation data, especially in two extreme cases: sample size is very small and exon numbers of two isoforms are quite different. Furthermore, by defining relevant quantitative measures (e.g., Jaccard correlation coefficient) and combining differential co-expression network analysis and GO functional enrichment analysis, a co-expression network analysis framework is developed to predict functions of isoforms and further, to discover their distinct functions within the same gene. We apply Iso-Net to study gene isoforms for several important transcription factors in human myeloid differentiation with the exon-level RNA-Seq data from three different cell lines. AVAILABILITY AND IMPLEMENTATION: Iso-Net is open source and freely available from https://github.com/Dingjie-Wang/Iso-Net.

Hepatitis B Virus Core Particles Containing a Conserved Region of the G Protein Combined with Interleukin-35 Protected Mice against Respiratory Syncytial Virus Infection without Vaccine-Enhanced Immunopathology.

Respiratory syncytial virus (RSV) is the most important cause of lower respiratory tract infection in infants and young children. The vaccine-enhanced disease (VED) has greatly hindered the development of an RSV vaccine. Currently, there are no licensed vaccines for RSV. In this study, immunization of mice with hepatitis B virus core particles containing a conserved region of the G protein (HBc-tG) combined with interleukin-35 (IL-35) elicited a Th1-biased response and a high frequency of regulatory T (Treg) cells and increased the levels of IL-10, transforming growth factor ß, and IL-35 production. Importantly, immunization with HBc-tG together with IL-35 protected mice against RSV infection without vaccine-enhanced immunopathology. To explore the mechanism of how IL-35 reduces lung inflammation at the gene expression level, transcription profiles were obtained from lung tissues of immunized mice after RSV infection by the Illumina sequencing technique and further analyzed by a systems biology method. In total, 2,644 differentially expressed genes (DEGs) were identified. Twelve high-influence modules (HIMs) were selected from these DEGs on the basis of the protein-protein interaction network. A detailed analysis of HIM10, involved in the immune response network, revealed that Il10 plays a key role in regulating the host response. The selected DEGs were consistently confirmed by quantitative real-time PCR (qRT-PCR). Our results demonstrate that IL-35 inhibits vaccine-enhanced immunopathology after RSV infection and has potential for development in novel therapeutic and prophylactic strategies.IMPORTANCE In the past few decades, respiratory syncytial virus (RSV) has still been a major health concern worldwide. The vaccine-enhance disease (VED) has hindered RSV vaccine development. A truncated hepatitis B virus core protein vaccine containing the conserved region (amino acids 144 to 204) of the RSV G protein (HBc-tG) had previously been shown to induce effective immune responses and confer protection against RSV infection in mice but to also lead to VED. In this study, we investigated the effect of IL-35 on the host response and immunopathology following RSV infection in vaccinated mice. Our results indicate that HBc-tG together with IL-35 elicited a balanced immune response and protected mice against RSV infection without vaccine-enhanced immunopathology. Applying a systems biology method, we identified Il10 to be the key regulator in reducing the excessive lung inflammation. Our study provides new insight into the function of IL-35 and its regulatory mechanism of VED at the network level.

scPADGRN: A preconditioned ADMM approach for reconstructing dynamic gene regulatory network using single-cell RNA sequencing data.

Disease development and cell differentiation both involve dynamic changes; therefore, the reconstruction of dynamic gene regulatory networks (DGRNs) is an important but difficult problem in systems biology. With recent technical advances in single-cell RNA sequencing (scRNA-seq), large volumes of scRNA-seq data are being obtained for various processes. However, most current methods of inferring DGRNs from bulk samples may not be suitable for scRNA-seq data. In this work, we present scPADGRN, a novel DGRN inference method using "time-series" scRNA-seq data. scPADGRN combines the preconditioned alternating direction method of multipliers with cell clustering for DGRN reconstruction. It exhibits advantages in accuracy, robustness and fast convergence. Moreover, a quantitative index called Differentiation Genes' Interaction Enrichment (DGIE) is presented to quantify the interaction enrichment of genes related to differentiation. From the DGIE scores of relevant subnetworks, we infer that the functions of embryonic stem (ES) cells are most active initially and may gradually fade over time. The communication strength of known contributing genes that facilitate cell differentiation increases from ES cells to terminally differentiated cells. We also identify several genes responsible for the changes in the DGIE scores occurring during cell differentiation based on three real single-cell datasets. Our results demonstrate that single-cell analyses based on network inference coupled with quantitative computations can reveal key transcriptional regulators involved in cell differentiation and disease development.

A general index for linear and nonlinear correlations for high dimensional genomic data.

BACKGROUND: With the advance of high throughput sequencing, high-dimensional data are generated. Detecting dependence/correlation between these datasets is becoming one of most important issues in multi-dimensional data integration and co-expression network construction. RNA-sequencing data is widely used to construct gene regulatory networks. Such networks could be more accurate when methylation data, copy number aberration data and other types of data are introduced. Consequently, a general index for detecting relationships between high-dimensional data is indispensable. RESULTS: We proposed a Kernel-Based RV-coefficient, named KBRV, for testing both linear and nonlinear correlation between two matrices by introducing kernel functions into RV2 (the modified RV-coefficient). Permutation test and other validation methods were used on simulated data to test the significance and rationality of KBRV. In order to demonstrate the advantages of KBRV in constructing gene regulatory networks, we applied this index on real datasets (ovarian cancer datasets and exon-level RNA-Seq data in human myeloid differentiation) to illustrate its superiority over vector correlation. CONCLUSIONS: We concluded that KBRV is an efficient index for detecting both linear and nonlinear relationships in high dimensional data. The correlation method for high dimensional data has possible applications in the construction of gene regulatory network.

Control of multilayer biological networks and applied to target identification of complex diseases.

BACKGROUND: Networks have been widely used to model the structures of various biological systems. The ultimate aim of research on biological networks is to steer biological system structures to desired states by manipulating signals. Despite great advances in the linear control of single-layer networks, it has been observed that many complex biological systems have a multilayer networked structure and extremely complicated nonlinear processes. RESULT: In this study, we propose a general framework for controlling nonlinear dynamical systems with multilayer networked structures by formulating the problem as a minimum union optimization problem. In particular, we offer a novel approach for identifying the minimal driver nodes that can steer a multilayered nonlinear dynamical system toward any desired dynamical attractor. Three disease-related biology multilayer networks are used to demonstrate the effectiveness of our approaches. Moreover, in the set of minimum driver nodes identified by the algorithm we proposed, we confirmed that some nodes can act as drug targets in the biological experiments. Other nodes have not been reported as drug targets; however, they are also involved in important biological processes from existing literature. CONCLUSIONS: The proposed method could be a promising tool for determining higher drug target enrichment or more meaningful steering nodes for studying complex diseases.

A New Model System for Exploring Assembly Mechanisms of the HIV-1 Immature Capsid In Vivo.

The assembly of the HIV-1 immature capsid (HIC) is an essential step in the virus life cycle. In vivo, the HIC is composed of [Formula: see text] hexameric building blocks, and it takes 5-6 min to complete the assembly process. The involvement of numerous building blocks and the rapid timecourse makes it difficult to understand the HIC assembly process. In this work, we study HIC assembly in vivo by using differential equations. We first obtain a full model with 420 differential equations. Then, we reduce six addition reactions for separate building blocks to a single complex reaction. This strategy reduces the full model to 70 equations. Subsequently, the theoretical analysis of the reduced model shows that it might not be an effective way to decrease the HIC concentration at the equilibrium state by decreasing the microscopic on-rate constants. Based on experimental data, we estimate that the nucleating structure is much smaller than the HIC. We also estimate that the microscopic on-rate constant for nucleation reactions is far less than that for elongation reactions. The parametric collinearity investigation testifies the reliability of these two characteristics, which might explain why free building blocks do not readily polymerize into higher-order polymers until their concentration reaches a threshold value. These results can provide further insight into the assembly mechanisms of the HIC in vivo.

Mathematical modeling and quantitative analysis of HIV-1 Gag trafficking and polymerization.

Gag, as the major structural protein of HIV-1, is necessary for the assembly of the HIV-1 sphere shell. An in-depth understanding of its trafficking and polymerization is important for gaining further insights into the mechanisms of HIV-1 replication and the design of antiviral drugs. We developed a mathematical model to simulate two biophysical processes, specifically Gag monomer and dimer transport in the cytoplasm and the polymerization of monomers to form a hexamer underneath the plasma membrane. Using experimental data, an optimization approach was utilized to identify the model parameters, and the identifiability and sensitivity of these parameters were then analyzed. Using our model, we analyzed the weight of the pathways involved in the polymerization reactions and concluded that the predominant pathways for the formation of a hexamer might be the polymerization of two monomers to form a dimer, the polymerization of a dimer and a monomer to form a trimer, and the polymerization of two trimers to form a hexamer. We then deduced that the dimer and trimer intermediates might be crucial in hexamer formation. We also explored four theoretical combined methods for Gag suppression, and hypothesized that the N-terminal glycine residue of the MA domain of Gag might be a promising drug target. This work serves as a guide for future theoretical and experimental efforts aiming to understand HIV-1 Gag trafficking and polymerization, and might help accelerate the efficiency of anti-AIDS drug design.

Identifying key nodes in multilayer networks based on tensor decomposition.

The identification of essential agents in multilayer networks characterized by different types of interactions is a crucial and challenging topic, one that is essential for understanding the topological structure and dynamic processes of multilayer networks. In this paper, we use the fourth-order tensor to represent multilayer networks and propose a novel method to identify essential nodes based on CANDECOMP/PARAFAC (CP) tensor decomposition, referred to as the EDCPTD centrality. This method is based on the perspective of multilayer networked structures, which integrate the information of edges among nodes and links between different layers to quantify the importance of nodes in multilayer networks. Three real-world multilayer biological networks are used to evaluate the performance of the EDCPTD centrality. The bar chart and ROC curves of these multilayer networks indicate that the proposed approach is a good alternative index to identify real important nodes. Meanwhile, by comparing the behavior of both the proposed method and the aggregated single-layer methods, we demonstrate that neglecting the multiple relationships between nodes may lead to incorrect identification of the most versatile nodes. Furthermore, the Gene Ontology functional annotation demonstrates that the identified top nodes based on the proposed approach play a significant role in many vital biological processes. Finally, we have implemented many centrality methods of multilayer networks (including our method and the published methods) and created a visual software based on the MATLAB GUI, called ENMNFinder, which can be used by other researchers.

Dynamic analysis of a drug resistance evolution model with nonlinear immune response.

Recent studies have utilized evolutionary mechanisms to impede the emergence of drug-resistant populations. In this paper, we develop a mathematical model that integrates hormonal treatment, immunotherapy, and the interactions among three cell types: drug-sensitive cancer cells, drug-resistant cancer cells and immune effector cells. Dynamical analysis is performed, examining the existence and stability of equilibria, thereby confirming the model's interpretability. Model parameters are calibrated using available prostate cancer data and literature. Through bifurcation analysis for drug sensitivity under different immune effector cells recruitment responses, we find that resistant cancer cells grow rapidly under weak recruitment response, maintain at a low level under strong recruitment response, and both may occur under moderate recruitment response. To quantify the competitiveness of sensitive and resistant cells, we introduce the comprehensive measures R1 and R2, respectively, which determine the outcome of competition. Additionally, we introduce the quantitative indicators CIE1 and CIE2 as comprehensive measures of the immune effects on sensitive and resistant cancer cells, respectively. These two indicators determine whether the corresponding cancer cells can maintain at a low level. Our work shows that the immune system is an important factor affecting the evolution of drug resistance and provides insights into how to enhance immune response to control resistance.

Identification of the molecular mechanisms for cell-fate selection in budding yeast through mathematical modeling.

The specification and maintenance of cell fates is essential to the development of multicellular organisms. However, the precise molecular mechanisms in cell fate selection are, to our knowledge, poorly understood due to the complexity of multiple interconnected pathways. In this study, model-based quantitative analysis is used to explore how to maintain distinguished cell fates between cell-cycle commitment and mating arrest in budding yeast. We develop a full mathematical model of an interlinked regulatory network based on the available experimental data. By theoretically defining the Start transition point, the model is able to reproduce many experimental observations of the dynamical behaviors in wild-type cells as well as in Ste5-8A and Far1-S87A mutants. Furthermore, we demonstrate that a moderate ratio between Cln1/2→Far1 inhibition and Cln1/2→Ste5 inhibition is required to ensure a successful switch between different cell fates. We also show that the different ratios of the mutual Cln1/2 and Far1 inhibition determine the different cell fates. In addition, based on a new, definition of network entropy, we find that the Start point in wild-type cells coincides with the system's point of maximum entropy. This result indicates that Start is a transition point in the network entropy. Therefore, we theoretically explain the Start point from a network dynamics standpoint. Moreover, we analyze the biological bistablity of our model through bifurcation analysis. We find that the Cln1/2 and Cln3 production rates and the nonlinearity of SBF regulation on Cln1/2 production are potential determinants for irreversible entry into a new cell fate. Finally, the quantitative computations further reveal that high specificity and fidelity of the cell-cycle and mating pathways can guarantee specific cell-fate selection. These findings show that quantitative analysis and simulations with a mathematical model are useful tools for understanding the molecular mechanisms in cell-fate decisions.

Accurate prediction of glioma grades from radiomics using a multi-filter and multi-objective-based method.

Radiomics, providing quantitative data extracted from medical images, has emerged as a critical role in diagnosis and classification of diseases such as glioma. One main challenge is how to uncover key disease-relevant features from the large amount of extracted quantitative features. Many existing methods suffer from low accuracy or overfitting. We propose a new method, Multiple-Filter and Multi-Objective-based method (MFMO), to identify predictive and robust biomarkers for disease diagnosis and classification. This method combines a multi-filter feature extraction with a multi-objective optimization-based feature selection model, which identifies a small set of predictive radiomic biomarkers with less redundancy. Taking magnetic resonance imaging (MRI) images-based glioma grading as a case study, we identify 10 key radiomic biomarkers that can accurately distinguish low-grade glioma (LGG) from high-grade glioma (HGG) on both training and test datasets. Using these 10 signature features, the classification model reaches training Area Under the receiving operating characteristic Curve (AUC) of 0.96 and test AUC of 0.95, which shows superior performance over existing methods and previously identified biomarkers.

scLRTD : A Novel Low Rank Tensor Decomposition Method for Imputing Missing Values in Single-Cell Multi-Omics Sequencing Data.

With the successful application of single-cell sequencing technology, a large number of single-cell multi-omics sequencing (scMO-seq)data have been generated, which enables researchers to study heterogeneity between individual cells. One prominent problem in single-cell data analysis is the prevalence of dropouts, caused by failures in amplification during the experiments. It is necessary to develop effective approaches for imputing the missing values. Different with general methods imputing single type of single-cell data, we propose an imputation method called scLRTD, using low-rank tensor decomposition based on nuclear norm to impute scMO-seq data and single-cell RNA-sequencing (scRNA-seq)data with different stages, tissues or conditions. Furthermore, four sets of simulated and two sets of real scRNA-seq data from mouse embryonic stem cells and hepatocellular carcinoma, respectively, are used to carry out numerical experiments and compared with other six published methods. Error accuracy and clustering results demonstrate the effectiveness of proposed method. Moreover, we clearly identify two cell subpopulations after imputing the real scMO-seq data from hepatocellular carcinoma. Further, Gene Ontology identifies 7 genes in Bile secretion pathway, which is related to metabolism in hepatocellular carcinoma. The survival analysis using the database TCGA also show that two cell subpopulations after imputing have distinguished survival rates.

Mathematical modeling and dynamical analysis of anti-tumor drug dose-response.

Cancer is a serious threat to human health and life. Using anti-tumor drugs is one of the important ways for treating cancer. A large number of experiments have shown that the hormesis appeared in the dose-response relationship of various anti-tumor drugs. Modeling this phenomenon will contribute to finding the appropriate dose. However, few studies have used dynamical models to quantitatively explore the hormesis phenomenon in anti-tumor drug dose-response. In this study, we present a mathematical model and dynamical analysis to quantify hormesis of anti-tumor drugs and reveal the critical threshold of antibody dose. Firstly, a dynamical model is established to describe the interactions among tumor cells, natural killer cells and M2-polarized macrophages. Model parameters are fitted through the published experimental data. Secondly, the positivity of solution and bounded invariant set are given. The stability of equilibrium points is proved. Thirdly, through bifurcation analysis and numerical simulations, the hormesis phenomenon of low dose antibody promoting tumor growth and high dose antibody inhibiting tumor growth is revealed. Furthermore, we fit out the quantitative relationship of the dose-response of antibodies. Finally, the critical threshold point of antibody dose changing from promoting tumor growth to inhibiting tumor growth is obtained. These results can provide suggestions for the selection of appropriate drug dosage in the clinical treatment of cancer.

Application of interpretable machine learning for early prediction of prognosis in acute kidney injury.

Background: This study aimed to develop an algorithm using the explainable artificial intelligence (XAI) approaches for the early prediction of mortality in intensive care unit (ICU) patients with acute kidney injury (AKI). Methods: This study gathered clinical data with AKI patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) in the US between 2008 and 2019. All the data were further randomly divided into a training cohort and a validation cohort. Seven machine learning methods were used to develop the models for assessing in-hospital mortality. The optimal model was selected based on its accuracy and area under the curve (AUC). The SHapley Additive exPlanation (SHAP) values and Local Interpretable Model-Agnostic Explanations (LIME) algorithm were utilized to interpret the optimal model. Results: A total of 22,360 patients with AKI were finally enrolled in this study (median age, 69.5 years; female, 42.8%). They were randomly split into a training cohort (16770, 75%) and a validation cohort (5590, 25%). The eXtreme Gradient Boosting (XGBoost) model achieved the best performance with an AUC of 0.890. The SHAP values showed that Glasgow Coma Scale (GCS), blood urea nitrogen, cumulative urine output on Day 1 and age were the top 4 most important variables contributing to the XGBoost model. The LIME algorithm was used to explain the individualized predictions. Conclusions: Machine-learning models based on clinical features were developed and validated with great performance for the early prediction of a high risk of death in patients with AKI.