Cheng Zhang

5-Methylcytosine (5mC) plays an important role in gene regulation and development. Although nanopore sequencing has enabled direct detection of 5mC, existing methods still face several limitations, including poor generalization across species and sequence contexts (CpG/CHG/CHH), as well as suboptimal integration of sequence and current signals. Here, we present MethyNano, a deep learning framework incorporating a contrastive learning strategy to detect 5mC from nanopore reads. By encouraging more discriminative and stable representations, the contrastive objective improves the model’s sensitivity to rare sequence contexts and reduces its prediction uncertainty in challenging regions. Across datasets from A. thaliana, O. sativa, and H. sapiens, our model achieves superior performance on key metrics compared with other existing methods. Extensive cross-species and cross-motif experiments demonstrate the robust generalization performance of MethyNano, while dimensionality-reduction visualizations of learned features provide an intuitive view of the model’s efficient representation capability. Moreover, our ablation studies show that MethyNano’s architecture enables more effective integration of critical features, leading to higher predictive accuracy. The project code is available at https://github.com/baigeHUI/MethyNano and https://doi.org/10.5281/zenodo.19858400. Supplementary data are available at Bioinformatics online.

DNA is considered an ideal medium for constructing molecular circuits due to its high programmability and exceptional information density. Current DNA-based circuits are primarily constructed by using toehold-mediated strand displacement (TMSD) or enzyme-assisted reactions. While TMSD-based systems can perform logical operations, their functionality relies heavily on precise base pairing, resulting in complicated design processes and limited scalability. In contrast, enzyme-driven circuits offer a simplified design and support functional diversification. In this study, we developed a DNA polymerase-driven circuit that establishes a system-level programming framework for managing complex molecular states. By regulating strand binding, we defined three stable operational modes: OFF, ON, and the Blocked state (BLC). Based on this mechanism, programmable logic gates were constructed to achieve controllable computing operations. The system incorporates a modular input domain design, enabling dynamic switching among operational modes in response to different input signals: matched inputs specifically activate the target circuit, whereas mismatched inputs drive nontarget circuits into a blocked state. This programmability supports multilevel logical access control and path backtracking, mimicking computer directory mechanisms. Furthermore, this architecture enables conditional signal isolation and regulation, providing an innovative strategy for the hierarchical organization and state-aware management of complex molecular information systems.

Natural biological systems achieve precise cellular control through multidimensional signaling architectures that integrate sequence specificity, structural dynamics, and conformational switching. While synthetic DNA networks have been engineered primarily using sequence programmability, exclusive reliance on this dimension constrains signaling range and integrated regulation. Here, we report an allosteric DNA computing framework enabling the simultaneous integration of sequence programmability with conformational dynamics for integrated multilevel signal processing. By encoding conformational signals within polythymidine loops (0 to 40 nucleotides), this system executes loop-dependent logic operations with expanded signaling ranges. Moreover, catalytic allosteric hairpin assemblies achieve ~30-fold signal amplification with enhanced signal-to-noise ratios. Concurrently, allosteric DNA neural networks discriminate conformational signals based on loop lengths (7 to 15 nucleotides) at two-nucleotide resolution. Crucially, microRNA-responsive versions of this framework regulate gene expression, thereby bridging conformational signaling with genetic control regulations in vivo. Collectively, this work establishes a conformational signal-processing paradigm for adaptive DNA computing, paving the way for advanced synthetic biology and precision therapeutics.

Supporting information for manuscript

Nanoparticles (NP) spanning diverse materials and properties have the potential to encapsulate and to protect a wide range of therapeutic cargos to increase bioavailability, to prevent undesired degradation, and to mitigate toxicity. Fulvestrant, a selective estrogen receptor degrader, is commonly used for treating patients with estrogen receptor (ER)–positive breast cancer, but its broad and continual application is limited by poor solubility, invasive muscle administration, and drug resistance. Here, we developed an active targeting motif-modified, intravenously injectable, hydrophilic NP that encapsulates fulvestrant to facilitate its delivery via the bloodstream to tumors, improving bioavailability and systemic tolerability. In addition, the NP was coloaded with abemaciclib, an inhibitor of cyclin-dependent kinases 4 and 6 (CDK4/6), to prevent the development of drug resistance associated with long-term fulvestrant treatment. Targeting peptide modifications on the NP surface assisted in the site-specific release of the drugs to ensure specific toxicity in the tumor tissues and to spare normal tissue. The NP formulation (PPFA-cRGD) exhibited efficient tumor cell killing in both in vitro organoid models and in vivo orthotopic ER-positive breast cancer models without apparent adverse effects, as verified in mouse and Bama miniature pig models. This NP-based therapeutic provides an opportunity for continual and extensive clinical application of fulvestrant, thus indicating its promise as a treatment option for patients with ER-positive breast cancer.Significance:A smart nanomedicine encapsulating fulvestrant to improve its half-life, bioavailability, and tumor-targeting and coloaded with CDK4/6 inhibitor abemaciclib to block resistance is a safe and effective therapy for ER-positive breast cancer.

In biological systems, molecular network functionalities are usually switched in a flexible, facile, and programmable manner. Mimicking this, substantial studies are directed towards developing synthetic DNA networks that exhibit similar function-switching capabilities, though often hindered by extensive molecular architecture changes and stringent condition controls, which result in a time-consuming and labor-intensive process. Here, we develop a base stacking-mediated allostery strategy to manipulate the DNA computing function switching with minimal molecular architecture changes, usually as few as 1-2 nucleotide changes. We implement up to 20 distinct logic function switching within DNAzyme networks. We also validate our function switching platform to implement totally 84 kinds of gene regulation patterns in cancer cell lines, demonstrating its utility in RNA sensing and green fluorescent protein regulation. This strategy offers a simplified alternative approach to enrich DNA regulations, with potential applications in DNA computing and bioengineering.

Gold nanoparticles (AuNPs) have become a promising assembly material for exploring and understanding physicochemical mechanisms in vitro and in vivo. Recent reports of AuNP superstructure assembly methods with conditional manipulation features usually require specific physicochemical triggering conditions and macroscopic control of the whole reaction system, thus lacking simple and selective regulation of the assemblies of the AuNP superstructure. In this study, a DNAzyme‐based method is developed to program selectively controlled hierarchical assembly of AuNP superstructures. In which the DNAzyme‐catalyzed network enabled the selective manipulation of the generation of different AuNP superstructures by programming the activation of local subsystems. First, the programming manipulation of AuNP superstructure assembly is demonstrated using four kinds of circuit programs (YES, AND, OR, and Cascade). Then, a selectively controlled superstructure cascade assembly platform is established by merging two independent cascade subsystems, utilized to achieve on‐demand manipulation of the three superstructure assembly states. This work provides a simple and selective tool to simultaneously manipulate the hierarchical assembly of multiple nanoparticle superstructures for applications such as dynamic nanomachines and medical sensing.

DNA-based storage offers a compelling alternative to traditional optical and magnetic devices. However, random data access usually requires additional noncoding primer DNA as indexes, which substantially reduce the physical data density. Here, we propose an alternative strategy to overcome this barrier by loading different data-encoding DNA files into porous microspheres, each distinguished by unique photonic bandgaps and diameters, allowing for 10 ⁵ types of indexing. With the two features as the addressing indexes, the physical separation of subsets from a diverse pool of DNA files is achieved, thereby facilitating the selective retrieval of stored data. The interconnected nanopore arrays and uniformly distributed positive charges within the microspheres enhance DNA enrichment, achieving a storage density of up to 22.6 exabytes per gram, far exceeding that of previous random access storage methods. This work outlines a simple and scalable method for creating nonfading photonic indexes, enabling long-term random access while maintaining high storage capacity.