Molecular simulation techniques have become an invaluable tool for elucidating the fundamental principles of life at the molecular level. After nearly five decades of development, biomolecular simulations have evolved to enable the quantitative characterization of complex biomolecular events, such as protein folding, conformational dynamics, and protein–protein interactions. These advancements have significantly influenced both fundamental and applied research. In recent years, the integration of machine learning, particularly deep learning algorithms, has further driven innovation in this field. This perspective aims to discuss the latest advancements in biomolecular simulation techniques and to explore emerging applications, development trends, and major challenges in biomolecular dynamics simulations.

The traditional thinking has been that cancer cells require a great amount of glucose to support their rapid growth, but the reality may be different. We have previously demonstrated that all cancer cells in The Cancer Genome Atlas harbor persistent Fenton reactions in their cytosol, which generate {\text{OH}}^{-} and ultimately kill the cells by alkalosis if not neutralized timely. Here, we present data to show that (1) cancer cells uptake large amounts of glucose to produce sufficient levels of {\mathrm{H}}^{+}ions to keep the cytosolic pH stable, hence keeping the cells viable; (2) de novo nucleotide biosynthesis represents the predominant acidifying pathway and gets on average the largest allocation of glucose metabolic flux among the 19 cancer types investigated; and (3) although the {\mathrm{H}}^{+}ions produced by nucleotide biosynthesis and other acidifiers keep the cells alive, the synthesized nucleotides drive cancerous cell proliferation. Taken together, it is not that cancerous cell division requires high levels of glucose imports, instead it is the life‐saving nucleotide syntheses that drive cell division. Understanding this causal relationship correctly is significant since it explains why cancers depend so heavily on glucose but not on other nutrients. More importantly, this realization may lead to fundamentally novel and more effective ways to treat cancer.

Large‐scale artificial intelligence (AI) models can mine potential patterns from massive amounts of data and provide more accurate analyses. This capability has enabled its gradual application in various areas of bioinformatics. However, few reviews have comprehensively summarized the applications of different types of large‐scale AI models in key areas of bioinformatics. Therefore, we first introduce the concept of large‐scale AI models and classify them into three types. Second, we summarize the key methods, applications, and resources of these three types of bioinformatics models. Finally, we discuss challenges and directions for future research. This review provides researchers with a comprehensive perspective to better understand the applications of large‐scale AI models in bioinformatics.

Metabolic alterations and immune dysfunction within the gastric tumor microenvironment critically drive gastric cancer (GC) progression and therapeutic resistance. Although single‐cell RNA sequencing (scRNA‐seq) has unveiled cellular heterogeneity in GC, the metabolic landscapes of tumor cells and their interplay with immune components remain underexplored. By integrating scRNA‐seq data from 35,633 cells across 23 GC tissues (GSE150290), bulk RNA‐seq data from UCSC Xena, and two independent microarray cohorts (GSE26899, GSE62254), we systematically characterized metabolic heterogeneity and identified immune‐related prognostic biomarkers. Reclustering of malignant epithelial cells revealed distinct metabolic phenotypes, with the citrate cycle and oxidative phosphorylation pathways emerging as key drivers of intratumoral diversity and T cell differentiation. Through machine learning and survival analyses, we discovered a novel risk score model composed of 6 T cell differentiation signatures, which stratified patients into high‐ and low‐risk groups with significant differences in overall survival. Notably, this model outperformed traditional clinicopathological factors in predicting prognosis, validated in both bulk RNA‐seq and microarray datasets. Immunohistochemistry further confirmed the prognostic value of key regulatory proteins (RGS1, CXCR4, CTLA4, ARPP19, ZNRF1, and ZNF207). Our findings highlight the metabolic immune crosstalk in GC and provide a promising biomarker panel for precision risk stratification and potential immunotherapeutic targets.

Advances in deep learning and generative modeling have transformed the landscape of protein and RNA design, enabling rapid and precise creation of novel biomolecules with tailored structures and functions. In protein design, generative deep learning frameworks now support backbone generation, sequence optimization, and joint sequence–structure co‐design with unprecedented accuracy. These approaches have facilitated broad applications ranging from cyclic peptide and non‐natural fold engineering to functional tool development, including small‐molecule sensing, catalytic center scaffolding, allosteric switching, intracellular logic circuits, and the targeting of intrinsically disordered proteins. Emerging therapeutic applications—such as immune cell engineering, G protein‐coupled receptor‐targeted miniproteins, receptor‐degrading binders, and thermostable antitoxins—demonstrate the translational potential of computational design. Parallel progress in RNA design, driven by enhanced 3D structure prediction models and generative algorithms, is expanding capabilities in aptamer engineering and RNA–protein complexes, despite ongoing challenges in model generalization and experimental validation. Together, these developments highlight a new era of AI‐driven molecular engineering, in which unified protein–RNA modeling, large‐scale sampling, and automated experimental pipelines will accelerate the creation of programmable biological systems and next‐generation therapeutics.

Accurate color matching is essential to achieving aesthetically realistic outcomes in dental crown and bridge restorations. Traditional visual methods, however, are often affected by lighting variations and observer subjectivity. These limitations can lead to metamerism and inconsistent clinical outcomes. In this paper, we propose a deep learning‐based snapshot spectral reflectance prediction framework by incorporating multimodal optical priors into a physics‐informed modular reconstruction pipeline. Architecturally, it employs dual subnetworks integrated with attention mechanisms to enhance both spectral and spatial fidelity. A dedicated dataset consisting of over 4000 RGB–hyperspectral image pairs of dental samples is constructed under diverse lighting conditions using a custom‐built acquisition setup, enabling robust reflectance restoration. Quantitative evaluations demonstrate that the mean squared error of the reconstructed spectral reflectance is 0.002, with the structural similarity index measure achieving 0.8496. The proposed spectral reflectance reconstruction framework is fast, reliable, and robust, offering a physically consistent and clinically applicable solution for accurate color matching. Furthermore, its modular design provides flexibility, suggesting potential for adaptation and broader applicability across various biomedical imaging domains.

Computed tomography (CT) images are often severely corrupted by artifacts in the presence of metals. Existing supervised metal artifact reduction (MAR) approaches suffer from performance instability on known data due to their reliance on limited paired metal‐clean data, which limits their clinical applicability. Moreover, existing unsupervised methods face two main challenges: (1) The CT physical geometry is not effectively incorporated into the MAR process to ensure data fidelity; and (2) traditional heuristic regularization terms cannot fully capture the abundant prior knowledge. To overcome these shortcomings, we propose a diffusion model (DM)‐regularized implicit neural representation (INR) framework for CT image MAR. The INR integrates physical constraints and imposes data fidelity, whereas the pretrained DM provides prior knowledge to regularize the solution. Experimental results on both simulated and clinical data demonstrate the effectiveness and generalization ability of our method, highlighting its potential to be applied in clinical settings.

Anonymization is a crucial de‐identification technique that protects data privacy while ensuring its utility for model building. Current generative models such as generative adversarial networks and variational auto‐encoders (VAEs) have been applied to medical image anonymization but mainly focus on general image features, lacking specificity in regions of interest such as lesions. This study proposes a novel framework for brain magnetic resonance imaging anonymization, enabling the handling of lesion region prediction while preserving patient privacy. The framework consists of three stages: pre‐training VAEs to represent lesion and non‐lesion regions in latent space; fine‐tuning these latent representations using a diffusion model conditioned on spatial and temporal features; and generating medical image substitutions through joint decoding of lesion and non‐lesion latent representations. The comparative investigation has highlighted the benefits of our proposed methods, achieving a promising privacy–utility balance. In a small number of real sample scenarios, using synthetic samples with an 86% anonymity rate still enhanced the downstream segmentation task by 4.60% and the classification task by 8.75%. Our proposed framework offers significant improvements over existing methods in preserving privacy and maintaining data utility for lesion prediction tasks, which holds potential implications for enhanced privacy practices in medical imaging.

Deep learning methods, particularly exemplified by AlphaFold2, have revolutionized the field of protein structure prediction—an achievement recognized by the 2024 Nobel Prize in Chemistry awarded to its core developers. Despite this remarkable achievement, the broader protein folding problem is far from solved. Key challenges—each representing opportunities for future breakthroughs—include single‐sequence structure prediction, modeling protein dynamics, accurately predicting multimeric complexes, and effectively incorporating experimental constraints. Here we review recent progress in these key frontiers and share our perspective on future directions.

Bayes factor has advantages over p‐value as test statistics for association, particularly when comparing multiple non‐nested alternative models. An efficient method to compute Bayes factor for linear mixed model (LMM) in the context of genetic association studies is lacking. In this study, we transform the standard LMM to a Bayesian linear regression by substituting the random effect with fixed effects, where the covariates of the fixed effects are eigenvectors of the genetic relatedness matrix and their respective prior effect sizes are proportional to the corresponding eigenvalues. Using conjugate normal inverse gamma priors on regression parameters, Bayes factors can be computed in a closed form. We demonstrate numerically the known relationship between Bayes factors and p‐values for the LMM. We then show that predictions based on the transformed Bayesian linear regression are identical to those of the best linear unbiased prediction (BLUP) of the standard LMM. Our results provided a new perspective and derivation to a known connection between BLUP and Bayesian estimates. Methods described in this note are implemented in the software IDUL as two new functionalities: computing Bayes factors and residuals for the LMM.