Spatial and Temporal Mapping of Breast Cancer Lung Metastases Identify TREM2 Macrophages as Regulators of the Metastatic Boundary.

Cancer mortality primarily stems from metastatic recurrence, emphasizing the urgent need for developing effective metastasis-targeted immunotherapies. To better understand the cellular and molecular events shaping metastatic niches, we used a spontaneous breast cancer lung metastasis model to create a single-cell atlas spanning different metastatic stages and regions. We found that premetastatic lungs are infiltrated by inflammatory neutrophils and monocytes, followed by the accumulation of suppressive macrophages with the emergence of metastases. Spatial profiling revealed that metastasis-associated immune cells were present in the metastasis core, with the exception of TREM2+ regulatory macrophages uniquely enriched at the metastatic invasive margin, consistent across both murine models and human patient samples. These regulatory macrophages (Mreg) contribute to the formation of an immune-suppressive niche, cloaking tumor cells from immune surveillance. Our study provides a compendium of immune cell dynamics across metastatic stages and niches, informing the development of metastasis-targeting immunotherapies. SIGNIFICANCE: Temporal and spatial single-cell analysis of metastasis stages revealed new players in modulating immune surveillance and suppression. Our study highlights distinct populations of TREM2 macrophages as modulators of the microenvironment in metastasis, and as the key immune determinant defining metastatic niches, pointing to myeloid checkpoints to improve therapeutic strategies. This article is featured in Selected Articles from This Issue, p. 2489.

Artificial immune system features added to breast cancer clinical data for machine learning (ML) applications.

We here propose a new method of combining a mathematical model that describes a chemotherapy treatment for breast cancer with a machine-learning (ML) algorithm to increase performance in predicting tumor size using a five-step procedure. The first step involves modeling the chemotherapy treatment protocol using an analytical function. In the second step, the ML algorithm is trained to predict the tumor size based on clinico-pathological data and data obtained from magnetic resonance imaging results at different time points of treatment. In the third step, the model is solved according to adjustments made at the individual patient level based on the initial tumor size. In the fourth step, the important variables are extracted from the mathematical model solutions and inserted as added features. In the final step, we applied various ML algorithms on the merged data. Performance comparison among algorithms showed that the root mean square error of the linear regression decreased with the addition of the mathematical results, and the accuracy of prediction as well as the F1-scores increased with the addition of the mathematical model to the neural network. We established these results for four different cohorts of women at different ages with breast cancer who received chemotherapy treatment.

Analysis of a breast cancer mathematical model by a new method to find an optimal protocol for HER2-positive cancer.

Treatment of breast cancer (positive for HER2, i.e., ERBB2) is described by a mathematical model involving non-linear ordinary differential equations with a hidden hierarchy. To reveal the hierarchy of dynamical variables of the system being considered, we applied the singular perturbed vector field (SPVF) method, where a system of equations can be decomposed to fast and slow sub-systems with explicit small parameters. This new form of the model, which is called a singular perturbed system, enables us to apply a semi-analytical method called the method of directly defining inverse mapping (MDDiM), which is based on the homotopy analysis asymptotic method. We introduced the treatment protocol in explicit form, through an analytical function that describes the exact dose and intervals between treatments in a cyclical manner. In addition, a new algorithm for the optimal dosage that causes tumour shrinkage is presented in this study. Furthermore, we took the concept of protocol optimisation a step further and derived a differential equation that represents vaccination depending on tumour size and yields an optimal protocol of different doses at every time point. We introduced the treatment protocol in explicit form, through an analytical function that describes the exact dose and intervals between treatments in a cyclical manner. In addition, a new algorithm for finding the optimal dosage that causes tumour shrinkage is presented in this study. Additionally, we took the concept of protocol optimisation a step further and derived a differential equation that represents vaccination depending on tumour size and yields an optimal protocol of different doses at every time point.

Machine learning vs. classic statistics for the prediction of IVF outcomes.

PURPOSE: To assess whether machine learning methods provide advantage over classic statistical modeling for the prediction of IVF outcomes. METHODS: The study population consisted of 136 women undergoing a fresh IVF cycle from January 2014 to August 2016 at a tertiary, university-affiliated medical center. We tested the ability of two machine learning algorithms, support vector machine (SVM) and artificial neural network (NN), vs. classic statistics (logistic regression) to predict IVF outcomes (number of oocytes retrieved, mature oocytes, top-quality embryos, positive beta-hCG, clinical pregnancies, and live births) based on age and BMI, with or without clinical data. RESULTS: Machine learning algorithms (SVM and NN) based on age, BMI, and clinical features yielded better performances in predicting number of oocytes retrieved, mature oocytes, fertilized oocytes, top-quality embryos, positive beta-hCG, clinical pregnancies, and live births, compared with logistic regression models. While accuracies were 0.69 to 0.9 and 0.45 to 0.77 for NN and SVM, respectively, they were 0.34 to 0.74 using logistic regression models. CONCLUSIONS: Our findings suggest that machine learning algorithms based on age, BMI, and clinical data have an advantage over logistic regression for the prediction of IVF outcomes and therefore can assist fertility specialists' counselling and their patients in adjusting the appropriate treatment strategy.

NeuroConstruct-based implementation of structured-light stimulated retinal circuitry.

BACKGROUND: Retinal circuitry provides a fundamental window to neural networks, featuring widely investigated visual phenomena ranging from direction selectivity to fast detection of approaching motion. As the divide between experimental and theoretical visual neuroscience is fading, neuronal modeling has proven to be important for retinal research. In neuronal modeling a delicate balance is maintained between bio-plausibility and model tractability, giving rise to myriad modeling frameworks. One biologically detailed framework for neuro modeling is NeuroConstruct, which facilitates the creation, visualization and analysis of neural networks in 3D. RESULTS: Here, we extended NeuroConstruct to support the generation of structured visual stimuli, to feature different synaptic dynamics, to allow for heterogeneous synapse distribution and to enable rule-based synaptic connectivity between cell populations. We utilized this framework to demonstrate a simulation of a dense plexus of biologically realistic and morphologically detailed starburst amacrine cells. The amacrine cells were connected to a ganglion cell and stimulated with expanding and collapsing rings of light. CONCLUSIONS: This framework provides a powerful toolset for the investigation of the yet elusive underlying mechanisms of retinal computations such as direction selectivity. Particularly, we showcased the way NeuroConstruct can be extended to support advanced field-specific neuro-modeling.

BCG and IL - 2 model for bladder cancer treatment with fast and slow dynamics based on SPVF method-stability analysis.

In this study, we apply the method of singularly perturbed vector field (SPVF) and its application to the problem of bladder cancer treatment that takes into account the combination of Bacillus CalmetteGurin vaccine (BCG) and interleukin (IL)-2 immunotherapy (IL - 2). The model is presented with a hidden hierarchy of time scale of the dynamical variables of the system. By applying the SPVF, we transform the model to SPS (Singular Perturbed System) form with explicit hierarchy, i.e., slow and fast sub-systems. The decomposition of the model to fast and slow subsystems, first of all, reduces significantly the time computer calculations as well as the long and complex algebraic expressions when investigating the full model. In addition, this decomposition allows us to explore only the fast subsystem without losing important biological/ mathematical information of the original system.The main results of the paper were that we obtained explicit expressions of the equilibrium points of the model and investigated the stability of these points.

Combination of singularly perturbed vector field method and method of directly defining the inverse mapping applied to complex ODE system prostate cancer model.

We propose a new method to solve a system of complex ordinary differential equations (ODEs) with hidden hierarchy. Given a complex system of the ODE, the hierarchy of the system is generally hidden. Once we reveal the hierarchy of the system, the system can be reduced into subsystems called slow and fast subsystems. This division of slow and fast subsystems reduces the analysis and hence reduces the computation time, which can be expensive. In our new method, we first apply the singularly perturbed vector field method that is the global quasi-linearization method. This method exposes the hierarchy of a given complex system. Subsequently, we apply a version of the homotopy analysis method called the method of directly defining the inverse mapping. We applied our new method to the immunotherapy of advanced prostate cancer.