Background: Metagenomics has become a powerful tool for understanding microbial communities in various environments. One key challenge is to accurately identify and group genomic fragments in different clusters, or "bins", especially when the reference genome is not available. Objective: To develop an improved approach for binning of metagenomics databy incorporating extra genomics signatures utilizing advanced deep clustering techniques and evaluating the proposed approach using suitable measures Methods: In this study, we employed advanced deep learning-based unsupervised methods to perform binning of Metagenomics data. Our approach encompassed the analysis of multiple datasets and incorporated several essential features, including Tetra Nucleotide Frequency (TNF), Oligonucleotide Frequency Derived Error Gradient (OFDEG), Coverage, length of contigs, and GC-Content. Further, we incorporated additional techniques for feature extraction and dimensionality reduction such as, Deep Embedded Clustering (DEC), Variational Autoencoder (VAE), and K-means clustering as a part of binning process. Results: The performance of the developed approach was compared to existing methods and tools using benchmarked low-complexity simulated and real metagenomics datasets. The approach was evaluated using appropriate measures, including the Silhouette index, Rand index, and Accuracy. Remarkably, the results demonstrated that our proposed approach achieved a Rand index exceeding 0.80, a Silhouette index above 0.60, and an accuracy surpassing 0.80. These results clearly indicate that our approaches outperformed all previous unsupervised-based methods and were on par with the semi-supervised binning approaches. Conclusion: In conclusion, this study presents a new insight for the binning of metagenomics data using the unsupervised deep learning-based approach, specifically designed to tackle the issue of limited reference data. The developed method exhibits highly promising results across many datasets, underscoring its potential as an innovative solution for metagenomics binning.

Industrial image anomaly detection (IAD) is crucial for quality control, but its performance often degrades when training data contain label noise. To circumvent the reliance on potentially flawed labels, unsupervised methods that learn from the data distribution itself have become a mainstream approach. Among various unsupervised techniques, student–teacher frameworks have emerged as a highly effective paradigm. Student–Teacher Feature Pyramid Matching (STPM) is a powerful method within this paradigm, yet it is susceptible to such noise. Inspired by STPM and aiming to solve this issue, this paper introduces Meta-Learning Guided STPM (MLG-STPM), a novel framework that enhances STPM’s robustness by incorporating a guidance mechanism inspired by meta-learning. This guidance is achieved through an Evolving Meta-Set (EMS), which dynamically maintains a small high-confidence subset of training samples identified by their low disagreement between student and teacher networks. By training the student network on a combination of the current batch and the EMS, MLG-STPM effectively mitigates the impact of noisy labels without requiring an external clean dataset or complex re-weighting schemes. Comprehensive experiments on the MVTec AD and VisA benchmark datasets with synthetic label noise (0% to 20%) demonstrate that MLG-STPM significantly improves anomaly detection and localization performance compared to the original STPM, especially under higher noise conditions, and achieves competitive results against other relevant approaches.

Objective.Pharmacokinetic (PK) parameters derived from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) provide quantitative characterization of tissue perfusion and permeability. However, existing deep learning methods for PK parameter estimation rely on either temporal or spatial features alone, overlooking the integrated spatial-temporal characteristics of DCE-MRI data. This study aims to remove this barrier by fully leveraging the spatial and temporal information to improve parameter estimation.Approach.A spatial-temporal information-driven unsupervised deep learning method (STUDE) was proposed. STUDE combines convolutional neural networks (CNNs) and a customized Vision Transformer to separately capture spatial and temporal features, enabling comprehensive modeling of contrast agent dynamics and tissue heterogeneity. Besides, a spatial-temporal attention feature fusion module was proposed to enable adaptive focus on both dimensions for more effective feature fusion. Moreover, the extended Tofts model imposed physical constraints on PK parameter estimation, enabling unsupervised training of STUDE. The accuracy and diagnostic value of STUDE was compared with the orthodox non-linear least squares (NLLS) and representative deep learning-based methods (i.e. gated recurrent unit, convolutional neural network, U-Net, and VTDCE-Net) on a numerical brain phantom and 87 glioma patients, respectively.Main results.On the numerical brain phantom, STUDE produced PK parameter maps with the lowest systematic and random errors even under low signal-to-noise ratio (SNR) conditions (SNR = 10 dB). On glioma data, STUDE generated parameter maps with reduced noise compared to NLLS and demonstrated superior structural clarity compared to other methods. Furthermore, STUDE outshined all other methods in the identification of glioma isocitrate dehydrogenase mutation status, achieving the area under the curve (AUC) values at 0.840 and 0.908 for the receiver operating characteristic curves ofKtransand Ve, respectively. A combination of all PK parameters improved AUC to 0.926.Significance.STUDE advances spatial-temporal information-driven and physics-informed learning for precise PK parameter estimation, demonstrating its potential clinical significance.

Differential analysis in proteomics is pivotal for biomarker discovery and disease mechanism elucidation, yet traditional statistical methods are constrained by distributional assumptions and empirical fold change threshold dependencies. This study systematically evaluates 18 unsupervised anomaly detection machine learning (ML) algorithms against the established statistical frameworks for differential protein detection from proteomic data sets. Using in silico simulated data sets derived from experimental data, we enabled cross-algorithm comparability through a probability based transformation. Results demonstrated that ML methods, particularly the Minimum Covariance Determinant (MCD), outperformed statistical test in recall, precision, and accuracy, with superior robustness to intersample heterogeneity. Validation on real-world proteomic data further confirmed that the MCD-identified differentially expressed proteins comprehensively covered canonical pathways while uncovering novel tumor-associated functional biomolecules. This work establishes unsupervised ML methods as robust alternatives to traditional hypothesis-driven statistical approaches in proteomics differential analysis, offering enhanced reliability for precision medicine research.

Abstract The Gaia mission’s third data release recorded low-resolution spectra for about 100 000 white dwarf candidates. A small subset of these spectra show evidence of characteristic broad Ca ii absorption features, implying the accretion of rocky material by so-called polluted white dwarfs – important probes of the composition of exoplanetary material. Several supervised and unsupervised data-intensive methods have recently been applied to identify polluted white dwarfs from the Gaia spectra. We present a comparison of these methods, along with the first application of t-distributed stochastic neighbour embedding (tSNE) to this dataset. We find that tSNE outperforms the similar technique Uniform Manifold Approximation and Projection (UMAP), isolating over 50 per cent more high-confidence polluted candidates, including 39 new candidates which are not selected by any other method investigated and which have not been observed at higher resolution. Supervised methods benefit greatly from data labels provided by earlier works, selecting many known polluted white dwarfs which are missed by unsupervised methods. Our work provides a useful case study in the selection of members of rare classes from a large, sporadically labelled dataset, with applications across astronomy.

Time series anomaly detection presents various challenges due to the sequential and dynamic nature of time-dependent data. Traditional unsupervised methods frequently encounter difficulties in generalization, often overfitting to known normal patterns observed during training and struggling to adapt to unseen normality. In response to this limitation, self-supervised techniques for time series have garnered attention as a potential solution to undertake this obstacle and enhance the performance of anomaly detectors. This paper presents a comprehensive review of the recent methods that make use of self-supervised learning for time series anomaly detection. A taxonomy is proposed to categorize these methods based on their primary characteristics, facilitating a clear understanding of their diversity within this field. The information contained in this survey, along with additional details that will be periodically updated, is available on the following GitHub repository: https://github.com/Aitorzan3/Awesome-Self-Supervised-Time-Series-Anomaly-Detection.

ABSTRACT Hyperspectral images are hindered by high dimensionality, making dimensionality reduction essential. Meanwhile, the ‘same objects but different spectra’ phenomenon complicates this process, as it obscures the true data structure and leads to poor low-dimensional representations. Additionally, the inherent redundancy in hyperspectral data further challenges model learning. While contrastive learning has achieved success in fields like computer vision, its application to hyperspectral images remains limited due to these issues. In this paper, an unsupervised dimensionality reduction method for hyperspectral images, two-layer contrastive learning based on hard positive samples, is proposed. At the sample-level, the k -hierarchical neighbours approach identifies hard positive samples (distant yet semantically similar points), which can be pulled closer together in the subspace. At the feature-level, a contrastive loss function is specially devised to enhance the discrimination ability and reduce redundancy. Experiments on four hyperspectral datasets demonstrate that our model outperforms existing dimensionality reduction techniques in classification accuracy, highlighting its effectiveness in addressing the unique challenges of hyperspectral image analysis.

Early detection of diseases is a critical pillar in advancing modern healthcare, offering timely interventions and better patient outcomes. This overview highlights a range of machine learning (ML) approaches that are transforming early disease diagnosis. We discuss how traditional supervised and unsupervised methods, alongside advanced deep learning and reinforcement learning techniques, are utilized to detect early disease markers, often before clinical symptoms appear. The paper begins with a discussion of ML fundamentals within healthcare, along with standard evaluation metrics such as accuracy, precision, recall, F1-score and AUC-ROC. It then explores various ML models, including supervised algorithms (support vector machines, decision trees and random forests), unsupervised methods (K-means, hierarchical clustering and principal component analysis) and deep learning architectures (convolutional neural networks, recurrent neural networks and transformers). Reinforcement learning's emerging role in healthcare is also examined. Practical applications across disease areas such as cancer, cardiovascular diseases, neurological disorders and infectious diseases are reviewed. We emphasize the importance of high-quality datasets, balanced data distribution and clinical relevance. Key challenges such as data scarcity, model interpretability, privacy, the risk of overdiagnosis and clinical integration are critically discussed. It underscores that the successful translation of these technologies from code to clinic hinges on a deep, bidirectional collaboration between data scientists and clinical experts to ensure that newly developed tools address real-world patient needs. The overview concludes with future directions, including explainable AI, federated learning, multimodal data fusion, real-time applications and quantum ML, charting the evolving path of early disease detection.

Evaluation of the diagnostic potential of Fourier transform-infrared spectroscopy on urine for urothelial bladder cancer: an in-hospital field study.

Synomaly noise and multi-stage diffusion: A novel approach for unsupervised anomaly detection in medical images.

An unsupervised bridge defect identification method based on multiscale masks and global awareness

Abstract Solar active regions (ARs) host the majority of solar eruptions. Studying the evolution and morphological features of ARs is significant for understanding the physical mechanisms of solar eruptions and beneficial for forecasting hazardous space weather. This work presents an automated DBSCAN-based solar active region detection (DSARD) method for ARs observed in magnetograms. DSARD is based on an unsupervised machine learning algorithm called density-based spatial clustering of applications with noise (DBSCAN). This method is employed to identify ARs in magnetograms observed by the Helioseismic and Magnetic Imager on board the Solar Dynamics Observatory from 2010 to 2023. To avoid duplicate detections and minimize projection effects, we focus on a longitudinal range of ±6° from the central meridian of the solar disk. Within this range, we obtain the distributions of the number, area, magnetic flux, tilt angle, and butterfly diagram of bipolar ARs in latitudes and time intervals during solar cycle 24, as well as their drift velocities. Most of these statistical results align with previous studies, which validates our method. The asymmetry indices of the number of ARs, cumulative area, and total unsigned magnetic flux indicate that the northern hemisphere dominated in terms of AR activity during most of solar cycle 24, except near solar maximum. Additionally, we analyze the dipole tilt angles of ARs in solar cycle 24 and the rising phase of solar cycle 25, revealing that 13% and 16% of ARs, respectively, violate Hale’s law.

Topological segmentation of mass spectrometry imaging data

Unsupervised domain adaptation method for bearing fault diagnosis assisted by twin data under extreme sample scarcity

An unsupervised fusion method for infrared and visible image under low-light condition based on Generative Adversarial Networks

Training a deep model for visual saliency detection requires the collection and labor-intensive annotation of overwhelmingly large data. We propose to learn saliency detection in a weakly supervised manner from single noisy label, which is easy to obtain from unsupervised handcrafted feature-based methods. However, deep networks tend to overfit such noises leading to a dramatic drop in accuracy. Given our goal, we address a natural question: can we identify outliers during network prediction and rectify the label noises? To this end, we propose a pixel-level noise mining framework for robust salient object detection (SOD) by exploiting its own knowledge, and without the need for external models. Specifically, during the early training stage, we progressively identify the outliers from a novel perspective during saliency detection, before the network overfits to the noisy labels, and generate a selection matrix in each iteration. Next, we adaptively rectify the label noises under the guidance of the selection matrix for better supervision in the later training stage. Extensive experiments on multiple benchmark datasets demonstrate the superiority of our method showing its ability to learn saliency detection comparable to state-of-the-art fully supervised methods. Furthermore, our approach outperforms existing weakly supervised methods utilizing single noisy label and surpasses the half of existing weakly supervised methods employing multiple noisy labels. Our approach, which trains with multiple noisy labels, outperforms all other methods employing multiple noisy labels across four major datasets. Furthermore, we also evaluate the generalization ability of our method on the multiclass semantic segmentation (SS) task. Our code is available at https://github.com/kendongdong/NoiseMining.

LEESDFormer: A lightweight unsupervised CNN-Transformer-based curve estimation network for low-light image enhancement, exposure suppression, and denoising.

A Bayesian Adaptive Clustered Prior Learning (ACPL) method for sparse spectroscopic regression.

Trichophyton indotineae is an emerging dermatophyte species responsible for recalcitrant and terbinafine-resistant dermatophytosis, raising concerns over diagnostic accuracy and treatment efficacy. This study aimed to improve the identification and resistance profiling of T. indotineae by integrating molecular methods with machine learning-assisted analysis of MALDI-TOF mass spectra. A total of 56 clinical isolates within the Trichophyton mentagrophytes complex were analyzed using ITS and ERG1 gene sequencing, antifungal susceptibility testing, and MALDI-TOF MS profiling. Terbinafine resistance was detected in 23 isolates and correlated with specific ERG1 mutations, including F397L, L393S, F415C, and A448T. While conventional MALDI-TOF MS failed to reliably distinguish T. indotineae from closely related species, unsupervised statistical methods (PCA and hierarchical clustering) revealed distinct spectral groupings. Supervised machine learning algorithms, particularly PLS-DA and SVM, achieved 100% balanced accuracy in species classification using 10-fold cross-validation. Biomarker analysis identified discriminatory spectral peaks for both T. indotineae and T. mentagrophytes (3417.29 m/z and 3423.53 m/z). These results demonstrate that combining MALDI-TOF MS with multivariate analysis and machine learning improves diagnostic resolution and may offer a practical alternative to sequencing in resource-limited settings. This approach could enhance the routine detection of terbinafine-resistant T. indotineae and support more targeted antifungal therapy.

The rapid integration of cryptocurrencies into the global financial ecosystem has introduced unprecedented challenges in market surveillance, risk management, and anomaly detection. While conventional statistical models such as ARIMA (Autoregressive Integrated Moving Average) and GARCH (Generalized Autoregressive Conditional Heteroscedasticity) have been widely used for anomaly detection, their reliance on assumptions of normality and stationarity often fails to capture the complexities of high-frequency, non-linear cryptocurrency trading. Furthermore, traditional risk metrics including down-to-up volatility, negative conditional skewness, and relative frequency may overlook short-term anomalies due to data aggregation limitations. In order to address these issues, this paper proposes machine-learning model for detecting anomalies in cryptocurrency markets using Jupyter Notebook. We compare four advanced unsupervised machine learning models, i.e, Density-Based Spatial Clustering of Applications with Noise (DBSCAN), Isolation Forest (iForest), One-Class Support Vector Machine (OC-SVM), and Local Outlier Factor (LOF) for anomaly detection by using Monte Carlo simulations. The findings indicate that DBSCAN has the highest precision (79.7%) with the fewest false positives, making it ideal for supervisory monitoring. However, the high false positive rates of OC-SVM and Isolation Forest limit their use. By using data of six well-known cryptocurrencies at three different temporal resolutions (daily, hourly, and 15-minute) the performance of these four unsupervised learning techniques also examined and confirmed that the anomalies identified by DBSCAN are also consistent with the other three methods. Additionally, for robustness of results, we use UpSet Plots to incorporate the shared anomalies and found across the three unsupervised learning methods. Number of anomalies also depends on the volatility and time interval of cryptocurrencies, more volatile / high frequency more anomalies. The study presents sound methodological approach for facilitating financial monitoring and mitigating risks in the cryptocurrencies market, and provides useful information for market players, analysts and policymakers. These results emphasize the importance of choosing algorithms based on specific surveillance targets to promote greater stability in digital asset environments.