Research Appraisals
Evidence-based critical appraisals of the latest medical research, systematically evaluated using Oxford CEBM methodology.
Showing 6 appraisals
Neural networks : the official journal of the International Neural Network Society
An approximation theory perspective on machine learning
A central problem in machine learning is often formulated as follows: Given a dataset [Formula: see text] , which is a sample drawn from an unknown probability distribution, the goal is to construct a functional model f such that f(x) ≈ y for any (x, y) drawn from the same distribution. Neural networks and kernel-based methods are commonly employed for this task due to their capacity for fast and parallel computation. The approximation capabilities, or expressive power, of these methods have been extensively studied over the past 35 years. In this paper, we will present examples of key ideas in this area found in the literature. We will discuss emerging trends in machine learning including the role of shallow/deep networks, approximation on manifolds, physics-informed neural surrogates, neural operators, and transformer architectures. Despite function approximation being a fundamental problem in machine learning, approximation theory does not play a central role in the theoretical foundations of the field. One unfortunate consequence of this disconnect is that it is often unclear how well trained models will generalize to unseen or unlabeled data. In this review, we examine some of the shortcomings of the current machine learning framework and explore the reasons for the gap between approximation theory and machine learning practice. We will then review some of recent work that achieves function approximation on unknown manifolds without the need to learn specific manifold features, such as the eigen-decomposition of the Laplace-Beltrami operator or atlas construction. In many machine learning problems, particularly classification tasks, the labels yj are drawn from a finite set of values. We summarize another recent paper that establishes a deep connection between signal separation problems and classification problems, proposing that classification tasks should be approached as instances of signal separation. We conclude by identifying several open research problems that warrant further investigation.
2 Aug 2026
Read appraisal →Proceedings of the National Academy of Sciences of the United States of America
Measuring disparate impact in human and machine decisions.
Empirical analyses have grown increasingly important in discrimination litigation with the greater availability of detailed data on individuals and decisions. A popular analytic strategy is to estimate disparities after adjusting for observed covariates, typically with a regression model, in hopes of ferreting out discriminatory intent. This approach, however, is ill-suited to auditing algorithms that are now commonly used to aid decisions, which typically do not include race or other legally protected factors as inputs. Motivated by legal understandings of disparate impact, we introduce an approach that aims to measure "unjustified" disparities in both human and machine decisions. Our method, which we call risk-adjusted regression, proceeds in three steps. In the first step, we combine all available information in a machine learning model to estimate the value, or inversely, the risk, of taking a certain action, such as approving a loan application or hiring a job candidate. Second, we measure disparities in decisions after adjusting for these risk estimates alone. Finally, in the third step, we assess the sensitivity of results to potential mismeasurement of risk. We demonstrate this approach on a detailed dataset of 2.2 million police stops of pedestrians in New York City, and show that traditional statistical tests of discrimination can substantially understate the magnitude of (risk-adjusted) racial disparities.
28 July 2026
Read appraisal →PloS one
Multi-view graph-regularized deep metric subspace clustering network
Multi-view subspace clustering has progressed significantly by using deep neural networks to handle nonlinear data representations. A recent advancement, the Multi-view Self-Expressive Subspace Clustering (MSESC) network, achieves markedly higher computational efficiency by substituting the traditional self-expression layer with a deep metric learning approach. Nevertheless, MSESC still suffers from two notable limitations: it fails to adequately capture the high-order geometric structures inherent in multi-view data, and it lacks effective guidance from the underlying clustering distribution. To overcome these shortcomings, we propose a novel framework termed Multi-View Graph Regularized Deep Metric Subspace Clustering (MVGR-DMSC). The proposed method introduces two key components into MSESC to enhance the discriminability of representations. First, a dual-order graph regularization module is devised to maintain both first-order and second-order manifold structures, thereby allowing the model to capture more complex local geometric relationships. Second, an adaptive view-weighted deep clustering module is incorporated, which employs the Kullback-Leibler divergence to guide representation learning while dynamically adjusting the contributions of different views. Through evaluations on five benchmark datasets, we show that MVGR-DMSC consistently yields better results than several state-of-the-art approaches, including the direct baseline MSESC, in both accuracy and robustness.
26 July 2026
Read appraisal →mSystems
MeLSI: Metric Learning for Statistical Inference in microbiome community composition analysis
Microbiome beta diversity analysis relies on distance-based methods, including permutational multivariate analysis of variance (PERMANOVA) combined with fixed ecological distance metrics (Bray-Curtis, Euclidean, Jaccard, and UniFrac), which treat all microbial taxa uniformly, regardless of their biological relevance to community differences. This "one-size-fits-all" approach may miss subtle but biologically meaningful patterns in complex microbiome data. We present Metric Learning for Statistical Inference (MeLSI), a novel machine learning framework that learns data-adaptive distance metrics optimized for detecting community composition differences in multivariate microbiome analyses. MeLSI employs an ensemble of weak learners using bootstrap sampling, feature subsampling, and gradient-based optimization to learn optimal feature weights, combined with rigorous permutation testing for statistical inference. The learned metrics can be used with PERMANOVA for hypothesis testing and with principal coordinates analysis for ordination visualization. Comprehensive validation on synthetic benchmarks and real data sets shows that MeLSI maintains proper type I error control while delivering competitive or superior statistical power for detecting subtle community shifts and, crucially, supplies interpretable feature-weight profiles that clarify which taxa drive group separation. On the DietSwap data set, MeLSI was the only method to achieve significance at α = 0.05, demonstrating that adaptive weighting can detect diet-induced community shifts that fixed metrics miss. Across all data sets, the learned feature weights identified biologically relevant taxa while providing actionable insight that no fixed distance metric can supply. MeLSI therefore offers a statistically rigorous tool that augments beta diversity analysis with transparent, data-driven interpretability.IMPORTANCEUnderstanding which microbes differ between groups of interest could reveal therapeutic targets and diagnostic biomarkers. However, current analysis methods treat all microbes equally (similar to using the same ruler to measure everything, regardless of what matters most). This means subtle but biologically important differences may go undetected, especially when only a few key species drive disease states while hundreds of "bystander" species add noise. Metric Learning for Statistical Inference (MeLSI) solves this by learning which microbes matter most for each specific comparison. In comparing male and female gut microbiomes, MeLSI identified specific bacterial families driving the differences, providing actionable biological insights that standard methods miss. This capability is particularly crucial for detecting early disease biomarkers, where differences are subtle and masked by biological variability. By telling researchers not just whether groups differ, but which specific microbes drive those differences, MeLSI accelerates the path from microbiome data to testable biological hypotheses and clinical applications.
22 July 2026
Read appraisal →American journal of human genetics
A transparent and generalizable deep-learning framework for genomic ancestry prediction
Accurately characterizing genetic ancestry is critical for ensuring reproducibility and fairness in genomic studies and downstream health research. This study aims to address the prediction of ancestry from genetic data using deep learning, with a focus on generalizability across datasets with diverse populations and on explainability to improve model transparency. We adapt the Diet Network, a deep-learning architecture proven to be effective in handling high-dimensional data, to learn population ancestry from single-nucleotide polymorphism (SNP) data using the populational Thousand Genomes Project dataset. Our results highlight the model's ability to generalize to diverse populations in the CARTaGENE, Montreal Heart Institute, and All of Us biobanks and that predictions remain robust to high levels of missing SNPs. We show that, despite the lack of North African populations in the training dataset, the model learns latent representations that reflect meaningful population structure for North African individuals in the biobanks. To improve model transparency, we apply Saliency Maps, DeepLift, GradientShap, and Integrated Gradients attribution techniques and evaluate their performance in identifying SNPs leveraged by the model. Using DeepLift, we show that the model's predictions are driven by population-specific signals consistent with those identified by traditional population-genetics metrics. This work presents a generalizable and interpretable deep-learning framework for genetic-ancestry inference in large-scale biobanks with genetic data. By enabling more widespread genomic ancestry characterization in these cohorts, this study contributes practical tools for integrating genetic data into downstream biomedical applications, supporting more inclusive and equitable healthcare solutions.
4 July 2026
Read appraisal →PloS one
Identifying policy-relevant traffic crash risk factors in Cheongju, South Korea using logistic regression and explainable machine learning
Rapid urbanization and increasing traffic volumes have made the occurrence of traffic crashes and the resulting harm a major public safety concern. This study analyzes traffic crash data from Cheongju, a mid-sized city in Chungcheongbuk-do, to identify key determinants of crash severity and provide evidence-based policy recommendations. Our approach is novel in that it integrates statistical modeling and machine learning methodologies; this dual approach not only overcomes the limitations inherent in using either technique alone but also allows for the identification of consistent risk factors influencing traffic crash severity that may have gone unrecognized otherwise. Marginal effects of explanatory variables were interpreted using ordinal logistic regression, while feature importance in machine learning models-including Support Vector Machine, Random Forest, XGBoost, and LightGBM-was evaluated using SHAP (SHapley Additive exPlanations) values. Both analytical approaches consistently identified traffic signal violations, failure to comply with safe driving obligations, and the absence of a median barrier on undivided roads as significant predictors of crash severity. By leveraging empirical data specific to Cheongju, our research provides regionally tailored insights that distinguish our work from prior studies with broader or less localized focus. These findings highlight the need for stricter enforcement of traffic regulations and structural improvements in roadway infrastructure and can inform policymakers in formulating effective, context-specific measures to enhance road safety.
24 June 2026
Read appraisal →