Patterns最新文献

Flow cytometry is a powerful technology for high-throughput protein quantification at the single-cell level. Technical advances have substantially increased data complexity, but novel bioinformatical tools often show limitations in statistical testing, data sharing, cross-experiment comparability, or clinical data integration. We developed MetaGate as a platform for interactive statistical analysis and visualization of manually gated high-dimensional cytometry data with integration of metadata. MetaGate provides a data reduction algorithm based on a combinatorial gating system that produces a small, portable, and standardized data file. This is subsequently used to produce figures and statistical analyses through a fast web-based user interface. We demonstrate the utility of MetaGate through a comprehensive mass cytometry analysis of peripheral blood immune cells from 28 patients with diffuse large B cell lymphoma along with 17 healthy controls. Through MetaGate analysis, our study identifies key immune cell population changes associated with disease progression.

The incidences of mental health illnesses, such as suicidal ideation and depression, are increasing, which highlights the urgent need for early detection methods. There is a growing interest in using natural language processing (NLP) models to analyze textual data from patients, but accessing patients’ data for research purposes can be challenging due to privacy concerns. Federated learning (FL) is a promising approach that can balance the need for centralized learning with data ownership sensitivity. In this study, we examine the effectiveness of FL models in detecting depression by using a simulated multilingual dataset. We analyzed social media posts in five different languages with varying sample sizes. Our findings indicate that FL achieves strong performance in most cases while maintaining clients’ privacy for both independent and non-independent client partitioning.

This paper argues that a range of current AI systems have learned how to deceive humans. We define deception as the systematic inducement of false beliefs in the pursuit of some outcome other than the truth. We first survey empirical examples of AI deception, discussing both special-use AI systems (including Meta’s CICERO) and general-purpose AI systems (including large language models). Next, we detail several risks from AI deception, such as fraud, election tampering, and losing control of AI. Finally, we outline several potential solutions: first, regulatory frameworks should subject AI systems that are capable of deception to robust risk-assessment requirements; second, policymakers should implement bot-or-not laws; and finally, policymakers should prioritize the funding of relevant research, including tools to detect AI deception and to make AI systems less deceptive. Policymakers, researchers, and the broader public should work proactively to prevent AI deception from destabilizing the shared foundations of our society.

Color is crucial in scientific visualization, yet it is often misused. Addressing this, we think accessible and accurate techniques, such as color-blind friendly palettes and perceptually even gradients, are vital. Accountability and basic knowledge in data visualization are key in fostering a culture of color integrity, ensuring accurate and inclusive data representation.

Deep-learning-based classification models are increasingly used for predicting molecular properties in drug development. However, traditional classification models using the Softmax function often give overconfident mispredictions for out-of-distribution samples, highlighting a critical lack of accurate uncertainty estimation. Such limitations can result in substantial costs and should be avoided during drug development. Inspired by advances in evidential deep learning and Posterior Network, we replaced the Softmax function with a normalizing flow to enhance the uncertainty estimation ability of the model in molecular property classification. The proposed strategy was evaluated across diverse scenarios, including simulated experiments based on a synthetic dataset, ADMET predictions, and ligand-based virtual screening. The results demonstrate that compared with the vanilla model, the proposed strategy effectively alleviates the problem of giving overconfident but incorrect predictions. Our findings support the promising application of evidential deep learning in drug development and offer a valuable framework for further research.

We present an end-to-end architecture for embodied exploration inspired by two biological computations: predictive coding and uncertainty minimization. The architecture can be applied to any exploration setting in a task-independent and intrinsically driven manner. We first demonstrate our approach in a maze navigation task and show that it can discover the underlying transition distributions and spatial features of the environment. Second, we apply our model to a more complex active vision task, whereby an agent actively samples its visual environment to gather information. We show that our model builds unsupervised representations through exploration that allow it to efficiently categorize visual scenes. We further show that using these representations for downstream classification leads to superior data efficiency and learning speed compared to other baselines while maintaining lower parameter complexity. Finally, the modular structure of our model facilitates interpretability, allowing us to probe its internal mechanisms and representations during exploration.

Structural neuroimaging studies have identified a combination of shared and disorder-specific patterns of gray matter (GM) deficits across psychiatric disorders. Pooling large data allows for examination of a possible common neuroanatomical basis that may identify a certain vulnerability for mental illness. Large-scale collaborative research is already facilitated by data repositories, institutionally supported databases, and data archives. However, these data-sharing methodologies can suffer from significant barriers. Federated approaches augment these approaches by enabling access or more sophisticated, shareable and scaled-up analyses of large-scale data. We examined GM alterations using Collaborative Informatics and Neuroimaging Suite Toolkit for Anonymous Computation, an open-source, decentralized analysis application. Through federated analysis of eight sites, we identified significant overlap in the GM patterns (n = 4,102) of individuals with schizophrenia, major depressive disorder, and autism spectrum disorder. These results show cortical and subcortical regions that may indicate a shared vulnerability to psychiatric disorders.

Spatially resolved transcriptomics has revolutionized genome-scale transcriptomic profiling by providing high-resolution characterization of transcriptional patterns. Here, we present our spatial transcriptomics analysis framework, MUSTANG (MUlti-sample Spatial Transcriptomics data ANalysis with cross-sample transcriptional similarity Guidance), which is capable of performing multi-sample spatial transcriptomics spot cellular deconvolution by allowing both cross-sample expression-based similarity information sharing as well as spatial correlation in gene expression patterns within samples. Experiments on a semi-synthetic spatial transcriptomics dataset and three real-world spatial transcriptomics datasets demonstrate the effectiveness of MUSTANG in revealing biological insights inherent in the cellular characterization of tissue samples under study.

In vitro fertilization (IVF) has revolutionized infertility treatment, benefiting millions of couples worldwide. However, current clinical practices for embryo selection rely heavily on visual inspection of morphology, which is highly variable and experience dependent. Here, we propose a comprehensive artificial intelligence (AI) system that can interpret embryo-developmental knowledge encoded in vast unlabeled multi-modal datasets and provide personalized embryo selection. This AI platform consists of a transformer-based network backbone named IVFormer and a self-supervised learning framework, VTCLR (visual-temporal contrastive learning of representations), for training multi-modal embryo representations pre-trained on large and unlabeled data. When evaluated on clinical scenarios covering the entire IVF cycle, our pre-trained AI model demonstrates accurate and reliable performance on euploidy ranking and live-birth occurrence prediction. For AI vs. physician for euploidy ranking, our model achieved superior performance across all score categories. The results demonstrate the potential of the AI system as a non-invasive, efficient, and cost-effective tool to improve embryo selection and IVF outcomes.

Treatment effect estimation (TEE) aims to identify the causal effects of treatments on important outcomes. Current machine-learning-based methods, mainly trained on labeled data for specific treatments or outcomes, can be sub-optimal with limited labeled data. In this article, we propose a new pre-training and fine-tuning framework, CURE (causal treatment effect estimation), for TEE from observational data. CURE is pre-trained on large-scale unlabeled patient data to learn representative contextual patient representations and fine-tuned on labeled patient data for TEE. We present a new sequence encoding approach for longitudinal patient data embedding both structure and time. Evaluated on four downstream TEE tasks, CURE outperforms the state-of-the-art methods, marking a 7% increase in area under the precision-recall curve and an 8% rise in the influence-function-based precision of estimating heterogeneous effects. Validation with four randomized clinical trials confirms its efficacy in producing trial conclusions, highlighting CURE’s capacity to supplement traditional clinical trials.