- Research Article
- 10.1016/j.jid.2025.10.579
Dynamic network biomarker analysis reveals predisease state in atopic dermatitis.
- May 01, 2026
- The Journal of investigative dermatology
- Eiryo Kawakami
Publications from 2021 to 2026
Showing 10 of 475 papers
Dynamic network biomarker analysis reveals predisease state in atopic dermatitis.
Uncertainty-aware variational soft prompt for few-shot tuning in fMRI-based mental state decoding
Functional magnetic resonance imaging (fMRI) is a widely leveraged signal in brain decoding, experiencing rapid advancement and demonstrating tremendous promises. The development of fMRI-based large models has piqued the interest of academics, who hope to overcome the existing limitations by taking advantage of advances in large model technology and increased accessibility of mental state datasets. This also raises the challenge of effectively transferring the pre-trained large models to new applications. To tackle this problem, this paper proposed a novel uncertainty-aware variational soft prompt approach to aid the pre-trained model in adapting to a new domain, while providing an estimation of the sample aleatoric uncertainty by using the reparameterization technique and associating the prompt vectors to uncertainty through the designed loss function. The comparative experiments showed that, the proposed approach can significantly improve model transfer performance and outperform the standard prompt method, in particular for dealing with few training samples. Moreover, the proposed approach demonstrated robustness to variations in model structure and the hyperparameter of the loss function, which also highlighted the practical usability of the proposed approach. The source code is publicly available at github.com/hzx-ntu/Uncertainty-aware-Variational-Soft-Prompt-for-Few-shot-Learning .
Read moreBandwidth adjustable Tanimoto kernel: a smooth alternative to the Gaussian kernel
Abstract Kernel methods have long played a central role in machine learning by enabling nonlinear modeling in high dimensional feature spaces. Among them, the Gaussian Radial Basis Function (RBF) kernel is widely used due to its smoothness, universality, and excellent performance in supervised learning. In contrast, the Jaccard/Tanimoto coefficient, originally defined over sets and binary vectors, has achieved notable success in chemoinformatics and information retrieval. Recent studies have shown that the Jaccard/Tanimoto coefficient can be extended to real-valued vectors and interpreted as a valid kernel. In this paper, we propose a natural generalization of the Tanimoto kernel by introducing a bandwidth-like parameter, leading to the Bandwidth-Adjustable Tanimoto (BAT) kernel. We focus on the shared structure of the BAT and the Gaussian RBF kernels in terms of their infinite polynomial expansions and investigate the universality of the BAT kernels, clarifying their relationship to the universal Taylor kernels. In particular, we show that by incorporating a bias term, the BAT kernel attains the universal approximation property comparable to that of the Gaussian kernel. Through experiments on classification, regression, and clustering tasks using benchmark datasets, we demonstrate that the BAT kernel performs comparably to the Gaussian kernel, while even the original Tanimoto kernel achieves competitive results. In particular, the BAT kernel attains performance on par with the Gaussian kernel in classification tasks, and it slightly surpasses the Gaussian kernel in regression and clustering tasks. A comparison of their functional forms, together with experiments on both regression and classification tasks involving nonlinear structures, further shows that the BAT kernel yields smoother predictive surfaces than the RBF kernel. Overall, our results indicate that the BAT kernel offers universality and practical performance, making it an alternative to the Gaussian kernel.
Read moreSequential Selling with Sunk Cost Bias
We study a sequential selling problem in which an agent receives daily offers to sell a good, incurs a holding cost each day, and is subject to sunk cost bias---allowing past, irrecoverable costs to influence present decisions. We introduce a formal model parameterizing the degree of sunk cost bias and distinguish between three behavioral types: optimistic (who ignore future bias), naive (who assume their current bias persists), and sophisticated (who anticipate the evolution of their own bias). For each type, we characterize the optimal selling strategy and precisely quantify the worst-case gap in expected objective profit compared to an unbiased agent. Our results show that optimistic agents can suffer a quadratic loss in profit due to excessive waiting, naive agents perform identically to unbiased agents, and sophisticated agents limit their losses to a linear function of the time horizon. These findings clarify how different anticipations of sunk cost bias affect sequential decision-making and suggest targeted interventions to mitigate inefficiency.
Read moreOn the Learning Dynamics of Two-layer Linear Networks with Label Noise SGD
One crucial factor behind the success of deep learning lies in the implicit bias induced by noise inherent in gradient-based training algorithms. Motivated by empirical observations that training with noisy labels improves model generalization, we delve into the underlying mechanisms behind stochastic gradient descent (SGD) with label noise. Focusing on a two-layer over-parameterized linear network, we analyze the learning dynamics of label noise SGD, unveiling a two-phase learning behavior. In Phase I, the magnitudes of model weights progressively diminish, and the model escapes the lazy regime; enters the rich regime. In Phase II, the alignment between model weights and the ground-truth interpolator increases, and the model eventually converges. Our analysis highlights the critical role of label noise in driving the transition from the lazy to the rich regime and minimally explains its empirical success. Furthermore, we extend these insights to Sharpness-Aware Minimization (SAM), showing that the principles governing label noise SGD also apply to broader optimization algorithms. Extensive experiments, conducted under both synthetic and real-world setups, strongly support our theory.
Read moreFair and Efficient Balanced Allocation for Indivisible Goods
We study the problem of allocating indivisible goods among agents with additive valuation functions to achieve both fairness and efficiency under the constraint that each agent receives exactly the same number of goods (the balanced constraint). While this constraint is common in real-world scenarios such as team drafts or asset division, it significantly complicates the search for allocations that are both fair and efficient. Envy-freeness up to one good (EF1) is a well-established fairness notion for indivisible goods. Pareto optimality (PO) and its stronger variant, fractional Pareto optimality (fPO), are widely accepted efficiency criteria. Our main contribution establishes both the existence and polynomial-time computability of allocations that are simultaneously EF1 and fPO under balanced constraints in two fundamental cases: (1) when agents have at most two distinct types of valuation functions, and (2) when each agent has a personalized bivalued valuation. Our algorithms leverage novel applications of maximum-weight matching in bipartite graphs and duality theory, providing the first polynomial-time solutions for these cases and offering new insights for constrained fair division problems.
Read moreMolecular Design with Artificial Intelligence: Progress and Perspectives for Small Molecules.
Progress in chemistry has been driven by the streamlining of inverse problem-solving methods. In the history of chemistry, several revolutionary technologies have led to leaps forward: the establishment of atomistic theory in the 19th century, structural analysis by spectroscopy in the 20th century, and the development of simulation by theoretical chemistry. Currently, chemistry is about to make a significant leap forward by integrating generative artificial intelligence (AI). In 2016, deep learning techniques were introduced in this domain, leading to explosive development. This paper reviews the development path, including traditional models such as variational autoencoders and more up-to-date models such as large language models and diffusion models. We also discuss how AI can have a real impact on chemistry, including the possibilities and problems associated with synthesizing AI-generated molecules.
Read moreImplementing generative artificial intelligence in precision oncology: safety, governance, and significance
The paramount challenge in precision oncology lies in further improving quality of life and response rates for individual patients. Efforts toward these goals are steadily expanding the scope of clinical implementation, despite ongoing challenges such as standardization, cost-effectiveness, and data harmonization. Building upon this maturing foundation, generative AI—which has evolved dramatically in recent years—is particularly valuable at this stage of advancing efficiency and adoption as an auxiliary technology linking literature, guidelines, trial protocols, and patient data. Specifically, through mutation interpretation, trial eligibility matching, and tumor board support, it is expected to contribute to advancing standardization, improving cost-effectiveness, accelerating data harmonization, and further accelerating human-centered decision-making. Accordingly, this review surveys the development history of generative AI and its current healthcare applications, organizing its implementation potential for precision oncology along three axes: (1) generative AI–based interpretation of genetic mutations and estimation of their pathological significance; (2) generative AI–driven verification of clinical trial eligibility; and (3) multimodal foundation models for imaging and pathology that compute “tumor phenotypes” using real-world data, contributing to report drafting and molecular surrogate estimation. In response, we propose a strategy centered on retrieval-augmented generation (RAG) and human-in-the-loop (HITL) workflows, encompassing data preparation based on OMOP, mCODE, and FHIR; multicenter prospective evaluation; auditable logs and governance aligned with Good Manufacturing Practice (GMP) and the EU AI Act; and a synthetic data strategy including differential privacy. Ultimately, this approach validates value through real-world outcomes and charts a path toward “learning oncology,” accelerating patient-centered decision-making and clinical trial development.
Read moreChemical genomics language model toward reliable and explainable compound-protein interaction exploration.
Accurate prediction of compound-protein interactions (CPIs) is crucial for chemical biology and drug discovery. Despite recent advancements, existing deep learning (DL)-based CPI models often struggle to simultaneously achieve high generalization performance, quantify prediction confidence, and ensure explainability. Here, we propose ChemGLaM, a chemical genomics language model designed to address these three crucial challenges, thereby enabling reliable and explainable CPI predictions. ChemGLaM integrates independently pre-trained chemical and protein language models through an interaction block with a cross-attention mechanism, achieving near state-of-the-art performance in predicting novel CPIs at a low computational cost. Incorporating uncertainty estimation and attention visualization enables ChemGLaM to enhance the success rate of virtual screening and to provide molecular insights into CPIs. To demonstrate the practical impact of ChemGLaM, we constructed a publicly available database containing large-scale CPI predictions for every possible pairing between all 20,434 human proteins and all 11,455 drugs and validated its practical applicability in a case study on amyotrophic lateral sclerosis. ChemGLaM marks an important step forward in addressing the challenges of AI-driven CPI exploration and drug discovery.Scientific ContributionThis study established a unified CPI prediction framework that simultaneously achieves high generalization performance, confidence quantification, and explainability. We leveraged this framework to create a community resource by constructing a comprehensive CPI database and demonstrated its practical utility by successfully prioritizing hit compounds and deconvoluting their targets in a phenotypic screening for amyotrophic lateral sclerosis.
Read moreCorrection to: Reaction, diffusion and non-local interaction.