Patterns最新文献

Data-driven machine learning, as a promising approach, possesses the capability to build high-quality, exact, and robust models from ophthalmic medical data. Ophthalmic medical data, however, presently exist across disparate data silos with privacy limitations, making centralized training challenging. While ophthalmologists may not specialize in machine learning and artificial intelligence (AI), considerable impediments arise in the associated realm of research. To address these issues, we design and develop FedEYE, a scalable and flexible end-to-end ophthalmic federated learning platform. During FedEYE design, we adhere to four fundamental design principles, ensuring that ophthalmologists can effortlessly create independent and federated AI research tasks. Benefiting from the design principles and architecture of FedEYE, it encloses numerous key features, including rich and customizable capabilities, separation of concerns, scalability, and flexible deployment. We also validated the applicability of FedEYE by employing several prevalent neural networks on ophthalmic disease image classification tasks.

Partially supervised segmentation is a label-saving method based on datasets with fractional classes labeled and intersectant. Its practical application in real-world medical scenarios is, however, hindered by privacy concerns and data heterogeneity. To address these issues without compromising privacy, federated partially supervised segmentation (FPSS) is formulated in this work. The primary challenges for FPSS are class heterogeneity and client drift. We propose a unified federated partially labeled segmentation (UFPS) framework to segment pixels within all classes for partially annotated datasets by training a comprehensive global model that avoids class collision. Our framework includes unified label learning (ULL) and sparse unified sharpness aware minimization (sUSAM) for class and feature space unification, respectively. Through empirical studies, we find that traditional methods in partially supervised segmentation and federated learning often struggle with class collision when combined. Our extensive experiments on real medical datasets demonstrate better deconflicting and generalization capabilities of UFPS.

In healthcare, machine learning (ML) shows significant potential to augment patient care, improve population health, and streamline healthcare workflows. Realizing its full potential is, however, often hampered by concerns about data privacy, diversity in data sources, and suboptimal utilization of different data modalities. This review studies the utility of cross-cohort cross-category (C⁴) integration in such contexts: the process of combining information from diverse datasets distributed across distinct, secure sites. We argue that C⁴ approaches could pave the way for ML models that are both holistic and widely applicable. This paper provides a comprehensive overview of C⁴ in health care, including its present stage, potential opportunities, and associated challenges.

Understanding tissue architecture and niche-specific microenvironments in spatially resolved transcriptomics (SRT) requires in situ annotation and labeling of cells. Effective spatial visualization of these data demands appropriate colorization of numerous cell types. However, current colorization frameworks often inadequately account for the spatial relationships between cell types. This results in perceptual ambiguity in neighboring cells of biological distinct types, particularly in complex environments such as brain or tumor. To address this, we introduce Spaco, a potent tool for spatially aware colorization. Spaco utilizes the Degree of Interlacement metric to construct a weighted graph that evaluates the spatial relationships among different cell types, refining color assignments. Furthermore, Spaco incorporates an adaptive palette selection approach to amplify chromatic distinctions. When benchmarked on four diverse datasets, Spaco outperforms existing solutions, capturing complex spatial relationships and boosting visual clarity. Spaco ensures broad accessibility by accommodating color vision deficiency and offering open-accessible code in both Python and R.

In a recent paper at Patterns, Hanchuan Peng, Peng Xie, and Feng Xiong from Southeast University describe a deep learning method to characterize complete single-neuron morphologies, which can discover neuron projection patterns of diverse cells and learn neuronal morphology representation. In this interview, the authors shared the story behind the paper and their research experience.

This interview is a companion to these authors’ recent paper, “DSM: Deep sequential model for complete neuronal morphology representation and feature extraction.”¹

Although several studies have deployed gradient boosting trees (GBT) as a robust classifier for federated learning tasks (federated GBT [FGBT]), even with dropout rates (federated gradient boosting trees with dropout rate [FDART]), none of them have investigated the overfitting effects of FGBT across heterogeneous and highly imbalanced datasets within federated environments nor the effect of dropouts in the loss function. In this work, we present the federated hybrid boosted forests (FHBF) algorithm, which incorporates a hybrid weight update approach to overcome ill-posed problems that arise from overfitting effects during the training across highly imbalanced datasets in the cloud. Eight case studies were conducted to stress the performance of FHBF against existing algorithms toward the development of robust AI models for lymphoma development across 18 European federated databases. Our results highlight the robustness of FHBF, yielding an average loss of 0.527 compared with FGBT (0.611) and FDART (0.584) with increased classification performance (0.938 sensitivity, 0.732 specificity).

Advancing precision oncology requires accurate prediction of treatment response and accessible prediction models. To this end, we present shinyDeepDR, a user-friendly implementation of our innovative deep learning model, DeepDR, for predicting anti-cancer drug sensitivity. The web tool makes DeepDR more accessible to researchers without extensive programming experience. Using shinyDeepDR, users can upload mutation and/or gene expression data from a cancer sample (cell line or tumor) and perform two main functions: "Find Drug," which predicts the sample’s response to 265 approved and investigational anti-cancer compounds, and "Find Sample," which searches for cell lines in the Cancer Cell Line Encyclopedia (CCLE) and tumors in The Cancer Genome Atlas (TCGA) with genomics profiles similar to those of the query sample to study potential effective treatments. shinyDeepDR provides an interactive interface to interpret prediction results and to investigate individual compounds. In conclusion, shinyDeepDR is an intuitive and free-to-use web tool for in silico anti-cancer drug screening.

Since Elon Musk’s purchase of Twitter/X and subsequent changes to that platform, computational social science researchers may be considering shifting their research programs to Mastodon and the fediverse. This article sounds several notes of caution about such a shift. We explain key differences between the fediverse and X, ultimately arguing that research must be with the fediverse, not on it.

Crosstalk among cells is vital for maintaining the biological function and intactness of systems. Most existing methods for investigating cell-cell communications are based on ligand-receptor (L-R) expression, and they focus on the study between two cells. Thus, the final communication inference results are particularly sensitive to the completeness and accuracy of the prior biological knowledge. Because existing L-R research focuses mainly on humans, most existing methods can only examine cell-cell communication for humans. As far as we know, there is currently no effective method to overcome this species limitation. Here, we propose MDIC3 (matrix decomposition to infer cell-cell communication), an unsupervised tool to investigate cell-cell communication in any species, and the results are not limited by specific L-R pairs or signaling pathways. By comparing it with existing methods for the inference of cell-cell communication, MDIC3 obtained better performance in both humans and mice.