Biology subjects

Yang, F.

Publications and source records attributed to Yang, F..

At least 37 records · Page 2Linked to original sources

bioRxiv2026

dbGIST: An LLM-Assisted Multi-Omics Resource for Target Exploration and Cross-Dataset Validation in Gastrointestinal Stromal Tumors

Gastrointestinal stromal tumors (GISTs) are the most common mesenchymal neoplasms of the gastrointestinal tract, yet GIST-specific omics evidence remains scattered across small cohorts and is not represented as a dedicated disease project in major cancer genomics resources, limiting reproducible target exploration. Here, we present dbGIST (https://www.dbgist.com), a dedicated GIST-focused multi-omics resource built to make dispersed GIST evidence searchable, analyzable, and reusable. dbGIST harmonizes data from 37 centers and 1,991 samples, including pathologically verified in-house cohorts, across genomics, bulk transcriptomics, proteomics, phosphoproteomics, and single-cell transcriptomics, and couples these data with curated clinical annotations covering survival, mutation status, risk stratification, metastasis or recurrence, mitotic index, tumor site and size, and imatinib response. The platform supports cohort-level molecular-clinical association, survival, enrichment, immune-infiltration, drug-sensitivity, and single-cell analyses through interactive visualizations, downloadable source data, and public APIs for programmatic access to reusable analysis outputs and visualization-ready data. An optional LLM-assisted interface helps users navigate analyses and interpret outputs. Using MCM7 as a case study, dbGIST linked a resource-derived candidate to survival, risk features, metastatic or recurrent disease, imatinib-response phenotypes, proliferative cell states, and in vitro GIST-cell behavior. dbGIST therefore provides a traceable and interoperable resource for target exploration and precision oncology research in GIST.

cancer biology↗

bioRxiv2026

Genos-m: a foundation model for human-associated microbial genomes

Human-associated microbial genomes encode extensive strain-level diversity and niche-specific gene repertoires that are critical to host health. However, these complex sequence features remain difficult to capture using general-purpose DNA foundation models, highlighting the need for dedicated representation learning tailored to the human microbiome. Here, we introduce Genos-m, an open-source foundation model for human-associated microbial genome representation. Genos-m was pretrained on approximately 1.2 trillion nucleotide tokens from a curated microbial genome corpus, including human-associated prokaryotic isolates, high-quality metagenome-assembled genomes (MAGs) and bacteriophages, supplemented with GTDB species-level representative genomes to broaden prokaryotic taxonomic breadth. The model uses a sparsely activated Mixture-of-Experts (MoE) Transformer architecture, with 4.7 billion total parameters, approximately 330 million activated parameters per forward pass and a maximum context length of one million base pairs. We evaluated frozen Genos-m representations across short-sequence and gene-level tasks, biosynthetic gene cluster (BGC)-based regional sequence tasks, whole-genome strain phenotype prediction, and zero-shot transfer on prokaryote-related RNAfitness assays. Across these benchmarks, Genos-m consistently ranked among the leading comparison models, with the best performance in five of eight gene-fitness regression tasks and in BGC type classification. Using sparse autoencoders, we identified sparse features in Genos-m hidden activations that aligned with annotated ORFs, intergenic regions, and tRNA and rRNA loci. In downstream applications, Genos-m-derived genome-informed species representations in-corporated into a human microbiome self-supervised learning model improved colorectal cancer (CRC)-control classification over conventional species-abundance random forest models. Genos-m also generated stable sample-level embeddings from as few as 10,000 metagenomic reads, retaining gut microbial community structure that distinguished geographic origin and aligned with enterotypes defined from full-depth taxonomic profiles. Together, these results support Genos-m as a reusable representation model for microbial genomes and metagenomes, with conclusions bounded by the reported datasets, task definitions and evaluation protocols. Genos-m model weights, inference code, and usage documentation are publicly available on GitHub (https://github.com/BGI-HangzhouAI/Genos-m) and Hugging-Face (https://huggingface.co/BGI-HangzhouAI/Genos-m).

bioinformatics↗

bioRxiv2026

Joint enzyme-reaction retrieval and catalytic optima prediction via multimodal fusion

MotivationEnzyme-reaction retrieval is increasingly used to prioritize candidate biocatalysts for experimental follow-up, where useful recommendations should indicate not only whether an enzyme can catalyze a target reaction but also under which pH and temperature conditions it should be tested. Existing retrieval models optimize catalytic matching scores, whereas catalytic optima predictors are typically developed as enzyme-level regressors because public pH and temperature annotations are sparse and often available only at the enzyme or EC-associated record level. This separation leaves a practical gap: high-ranking enzyme-reaction pairs are not evaluated for condition suitability, and enzyme-level optima predictions do not use the reaction context being retrieved. ResultsWe present GERO, a multimodal fusion framework that uses feature-gated cross-modal fusion to integrate global enzyme sequence semantics, sequence-derived pocket geometry, and molecular reaction representations for condition-aware enzyme-reaction retrieval and catalytic optima estimation with reaction context. To evaluate this setting, we define the tolerance-restricted hit rate (Hit@k-TR), which requires both top-k retrieval of the correct candidate and condition prediction within predefined tolerances. Across enzyme- and reaction-similarity splits, GERO improves Hit@k-TR over two-stage retrieval-then-prediction baselines. Representative benchmark examples and an iodinin biosynthesis case study further illustrate GEROs ability to provide candidate rankings together with plausible assay-condition estimates for downstream experimental prioritization. Availability and implementationSource code is available at https://github.com/ykxhs/GERO. Contactliujuan@whu.edu.cn Supplementary informationSupplementary data are available at XXXX online.

bioinformatics↗

bioRxiv2026

The Magnaporthe oryzae Effector AvrPii Attenuates Rice Immunity by Targeting the Calcium-Associated ANK Protein AVIN8

Rice blast, caused by the fungus Magnaporthe oryzae, poses a severe threat to global rice production. Although numerous M. oryzae effectors have been identified, the molecular mechanisms remain poorly understood. The effector AvrPii triggers ETI upon recognition by NLR receptor Pii and has been shown to target host exocytosis and oxidative metabolic pathways to suppress immunity. Here, we identify AVIN8 (AvrPii-interacting protein 8) as a new target of AvrPii. AVIN8 is an ankyrin-repeat (ANK) protein localized to the plasma membrane and nuclear envelope, with sequence similarity to known ANK calcium channels. AVIN8 expression is induced during early M. oryzae infection. Overexpression of AVIN8 enhances blast resistance, whereas its silencing increases susceptibility. AVIN8 promotes chitin-triggered ROS production in a Ca2+-dependent manner. In contrast, AvrPii impairs Ca2+ influx and chitin-triggered ROS burst. Our findings reveal a virulence strategy in which M. oryzae effector AvrPii hijacks the Ca2+ influx-associated ANK protein AVIN8 to suppress early calcium and PTI signaling in rice.

plant biology↗

bioRxiv2026

TDP-43 regulates chromatin looping and gene transcription through binding and stabilizing DNA G-quadruplex structures

TAR DNA-binding protein 43 (TDP-43) is a multifunctional DNA/RNA-binding protein implicated in transcriptional and post-transcriptional regulation. Dysregulation of TDP-43 is closely correlated with human diseases such as cancer and neurodegenerative diseases. Although its roles in RNA metabolism are well characterized, its function in transcriptional regulation remains largely underexplored. DNA G-quadruplexes (dG4s) are non-canonical nucleic acid structures enriched at gene promoters and regulatory elements, where they facilitate chromatin looping and gene transcription. Here, we investigated the transcriptional regulatory role of TDP-43 by integrating multi-omics datasets, including Hi-C, dG4 ChIP-seq, TDP-43 ChIP-seq, RNA-seq and ATAC-seq from K562 and HepG2 cells. Our analyses demonstrate TDP-43 binding and dG4s formation are highly colocalized at chromatin loop anchors, particularly at promoter and enhancer regions. TDP-43 occupancy at these anchors correlates with increased dG4 stability, chromatin loop interaction frequency, elevated chromatin accessibility, and upregulated gene expression. Morover, TDP-43 knockdown in HepG2 cells revealed a significant reduction in dG4 formation and loop interaction strength, accompanied by widespread transcriptional dysregulation. Collectively, our findings highlight a novel regulatory role of TDP-43 in facilitating long-range chromatin interactions and transcriptional activation through binding to and stabilizing dG4 structures, providing a mechanistic basis for gene dysregulation driven by TDP-43 dysfunction in diseases.

bioinformatics↗

bioRxiv2026

A unified population-genetic framework for inferring double reduction and higher-order two-locus linkage disequilibrium in autopolyploids

Polyploid genomes exhibit complex allelic interactions driven by polysomic inheritance that cannot be adequately captured by conventional pairwise linkage disequilibrium (LD) models, leaving higher-order dependency structures largely unresolved. Here, we develop a unified population-genetic framework that jointly infers double reduction and higher-order allelic associations in autopolyploids. By modeling dosage-dependent gametic structure, our approach reformulates LD as a hierarchical system, enabling decomposition of allelic dependencies beyond pairwise interactions. Through simulations, we show that conventional estimators systematically confound higher-order allelic structure with pairwise LD, leading to biased inference and loss of identifiability under dosage uncertainty. In contrast, our joint likelihood framework achieves asymptotically unbiased estimation and reveals that effective double reduction emerges as a composite parameter linking meiotic configuration to population-level allelic reshaping. Applying this model to Arabidopsis arenosa and cultivated potato, we uncover fundamental contrasts in genomic architecture. In natural populations, higher-order LD self-organizes into structured, long-range dependency networks, whereas predominantly clonal populations exhibit collapse of higher-order dependency structure. We further identify a spatial antagonism between double reduction and higher-order LD, demonstrating that elevated double reduction compresses the physical scale of allelic interactions. Our results demonstrate that autopolyploid genomes are organized as higher-order dependency systems rather than collections of pairwise associations, establishing higher-order allelic structure as a fundamental extension of classical LD theory.

genetics↗

bioRxiv2026

Adaptive response to long-term high temperatures during the reproductive development in Arabidopsis thaliana

Increasing global temperatures and the rising frequency of heat waves pose a significant threat to plant reproduction. The reproductive phase is particularly sensitive to heat stress, yet the underlying mechanisms regulating thermotolerance during this stage remain insufficiently understood, despite significant advcances in its understanding during vegetative growth. Heat stress responses are largely controlled by heat shock factors (HSFs) and their downstream targets, including heat shock proteins (HSPs). Among these, HSP101 is essential for acquired thermotolerance and recovery from stress, while HEAT SHOCK BINDING PROTEIN (HSBP) acts as a negative regulator of HSF activity, modulating the heat shock response. Here, we investigated the impact of elevated temperature regimes on the reproductive development of Arabidopsis thaliana, with a particular focus on pollen development and fertility. Our results show that heat stress negatively affects pollen development in a dose-dependent manner, leading to reduced reproductive success. We confirmed the critical role of HSP101 in reproductive thermotolerance using the hot1-3 mutant, deficient in HSP101. Furthermore, we provide evidence that the hot1-3 mutant is tetraploid. The origin of this event is unknown, but it is tempting to speculate that disruption of heat stress responses and interference with meiotic processes may lead to whole genome duplication. Overall, this study provides new insights into the regulation of plant reproductive development under heat stress and highlights the importance of HSP101 in maintaining fertility. These findings contribute to a better understanding of plant responses to rising temperatures and may inform strategies to enhance crop resilience under climate change. Main ConclusionFlowering Arabidopsis plants adapt to long-term high temperature by shortening the flowering period and reducing their fertility. The study also demonstrated that the commonly used hot1-3 mutant is tetraploid.

plant biology↗

bioRxiv2026

Fatty acid scavenging enables cancer escape from KRAS inhibition

Although inhibitors of oncogenic KRAS have shown clinical efficacy1, resistance to KRAS inhibition is common2, and its molecular basis remains unclear. Here we show that KRASi-resistant cancer cells sustain mitochondrial bioenergetics through enhanced fatty acid (FA) metabolism, despite suppression of canonical KRAS signaling. Specifically, KRASi-resistant pancreatic cancer cells exploit macropinocytosis to scavenge FA released from adipose tissue, fueling beta-oxidation independently of KRAS-PI3K signaling. This adaptive metabolic program is driven by the adhesion G protein-coupled receptor ADGRB1, which activates non-canonical PI3K{gamma}-PAK1 signaling to stimulate macropinocytosis and maintain metabolic homeostasis under KRASi. Disruption of ADGRB1-PI3K{gamma} signaling dismantles this metabolic program and restores KRASi sensitivity. This pathway operates across multiple KRAS-mutated cancers and is associated with poor therapeutic response and outcome. These findings offer a promising strategy for overcoming KRASi resistance.

cancer biology↗

bioRxiv2026

Inflammatory memory disables a DDR1 degradation checkpoint to enable pancreatic cancer growth

Inflammation and stroma remodeling regulate pancreatic ductal adenocarcinoma (PDAC), but how or if these cues are integrated at the molecular level remains unclear. Here, we identify a metabolic checkpoint that controls the stability of the collagen receptor DDR1, and subsequent tumorigenesis. We show that defective Col-I remodeling deprives PDAC cells of the high affinity DDR1 ligand, [3/4]Col-I, resulting in reduced ATP and activation of AMPK. AMPK phosphorylates DDR1 at T519, promoting its recognition by the E3 ubiquitin ligase adaptor FBXW2 and subsequent degradation. Importantly, this degradation pathway can be disabled by inflammatory signaling. Exposure to inflammatory cytokines induces methylation-dependent silencing of FBXW2, which establishes an inflammatory memory that preserves DDR1 stability, enabling sustained ligand-triggered receptor oligomerization and downstream NF-{kappa}B-NRF2 signaling even in restrictive stromal environments. Together, these findings identify regulated receptor turnover as a mechanism through which stromal architecture, metabolic state, and inflammatory memory are integrated to control PDAC progression.

cancer biology↗

bioRxiv2026

A novel subset of hepatocytes is simultaneously gluconeogenic and de novo lipogenic in the fed state and is naturally insulin resistant

It is generally accepted that hepatic gluconeogenesis, the synthesis of glucose from non-carbohydrate substrates is active in the fasted state and inactive in the fed state. In contrast, de novo lipogenesis is active in the fed state and is inactive in the fasted state. Here, we used targeted single cell RNA-seq, HCR RNA-FISH, and PrimeFlow in normal physiological mouse liver, and identified a subpopulation of periportal hepatocytes that simultaneously co-express both gluconeogenic and lipogenic genes in the fed state. Euglycemic-hyperinsulinemic clamps further demonstrated that this novel hepatocyte subpopulation is naturally insulin resistant. Spatial metabolic imaging coupled with stable isotope tracing analyses revealed individual hepatocytes that simultaneously undergo both gluconeogenesis and de novo lipogenesis. These dual-positive hepatocytes were also present in human hepatocytes from humanized mouse livers. Moreover, the number of dual-positive hepatocytes increased in high-fat diet-fed mice, suggesting a paradigm shift in our understanding of how the liver becomes insulin resistant.

molecular biology↗

bioRxiv2026

Structural basis for repurposing a flexible phage tail into an Intraspecific bacterial competition weapon

The rise of multidrug-resistant bacteria poses a major threat to human health. Tailocins, which are phage tail-like bacteriocins, offer a way to kill specific bacteria without the risks of horizontal gene transfer found in phage therapy. While the mechanism of contractile R-type tailocins is well-established, the bactericidal action of non-contractile F-type tailocins remains unresolved. Here, we report the high-resolution cryo-EM structure of the F-pyocin from Pseudomonas aeruginosa. The structure reveals a complex nanomachine made of a terminator cap (AlpD/PA0910), a helical tube (PA0633), and a tail tip that mediates a symmetry transition from sixfold to threefold. Notably, the tail tip connects to an elongated central fiber (PA0641) decorated with trimeric side fibers (PA0643/PA0646) involved in host recognition. Through genetic deletion studies and proteomic analysis, we delineate the F-pyocin assembly pathway, establishing that the tape-measure protein (PA0636) serves as a central scaffold coordinating the assembly of the tube, cap, and tail-tip modules. Structural analysis reveals a localized register shift in the central fibers coiled-coil shaft, suggesting a metastable state that may facilitate conformational changes upon host recognition. Our findings detail the atomic structure of an F-type tailocin, uncovering its coordinated assembly pathway and specialized tail-fiber architecture, which establishes a structural framework for engineering next-generation, precision antimicrobials.

biophysics↗

bioRxiv2026

Stemona genomes illuminate fatty acid partitioning between seeds and elaiosomes mediating wasp dispersal

O_LISeed dispersal by wasps (vespicochory) is documented in five angiosperm families, with phylogenies suggesting that most vespicochorous lineages evolved from ant-dispersed (myrmecochorous) ancestors. While recent work has identified cues attracting wasp dispersers, the molecular basis remains unclear. C_LIO_LITo shed light on the molecular basis of vespicochory, we generated chromosome-level genomes for the wasp-dispersed Stemona tuberosa, and its close ant-dispersed relative S. mairei. Combining comparative genomic, transcriptomic, lipidomic, and functional analyses, we ask (i) how chemical elaiosome differentiation occurs during seed development, and (ii) how key mutualism-associated genes evolved in Stemona. C_LIO_LIWe show that elaiosomes up-regulate stearoyl-ACP {Delta}9 desaturases (SAD) and accumulate oleic acid and 1,2-diolein, which serve both as food rewards and as precursors for CER1/3-mediated biosynthesis of (Z)-9-tricosene, the key wasp attractant. By contrast, high expression of FatB and DGAT1 in seeds is associated with medium-chain fatty acids (MCFAs) production. Comparative genomics indicates conservation of these genes across Stemona, with S. tuberosa-specific FatB expansion and SAD/CER1/3 repertoire divergence. C_LIO_LIOur work supports a model for elaiosome-seed fatty-acid differentiation linking recruitment and nourishment in vespicochory. Given that oleic acid and 1,2-diolein are also key cues in myrmecochory, this shared biochemical basis may help explain how vespicochory evolves from ant-dispersed ancestors. C_LI

evolutionary biology↗

bioRxiv2026

Mutant FGFR3 restricts bone yet expands cortex via ERK-mediated self-repression

FGFR3 gain-of-function mutations cause achondroplasia 1,2, the most common form of dwarfism, yet trigger paradoxical cerebral overgrowth and skeletal stunting 3-9. Here, we demonstrate that Fgfr3-K644E mutant drives enhanced ERK activation, which then represses Fgfr3 expression and downregulates ERK signaling. In the developing cortex, this mechanism transiently expands cortical stem cells by activating ERK, then shifts signaling to downregulate Fgfr3/ERK/PKA and enhance YAP/TAZ, promoting cerebral overgrowth and premature ependymal maturation. In growing bones, Fgfr3-K644E mutant first elevates ERK to inhibit chondrocyte pre-hypertrophy, then downregulates Fgfr3/ERK/PKA and upregulates YAP/TAZ to accelerate premature hypertrophy and ossification. Our findings establish that FGFR3 mutations limit bone yet expand cortical size via an ERK negative feedback loop, revealing a unified disease mechanism and novel therapeutic targets.

developmental biology↗

bioRxiv2026

m6A-dependent microRNA binding to chromatin-associated RNA for transcriptional activation

For decades, microRNAs (miRNAs) have been canonically viewed as post-transcriptional repressors. We discovered extensive binding of microRNAs to chromatin-associated RNAs (caRNAs) and uncovered an N6-methyladenosine (m6A)-dependent transcriptional activation mechanism of microRNAs. We show that m6A-binding proteins FXR1/2 anchor AGO1/2 at m6A-marked caRNAs, where AAGUGC-seed microRNAs function as guide RNAs to direct AGO positioning. This dual anchoring stabilizes the AGO-microRNA/FXR-m6A complex at specific loci, which in turn recruits the ATP-dependent chromatin remodeler SMARCA4 (BRG1) to promote local chromatin opening and TET1 for DNA demethylation, respectively. Together, these coordinated activities establish a transcriptionally permissive chromatin environment, enhancing accessibility and transcription across hundreds of genes in diverse cell types. Beyond the AAGUGC-seed family, additional microRNAs and siRNAs also enhance transcription, suggesting that caRNA binding and transcriptional activation may represent a broader property of small RNAs.

molecular biology↗

bioRxiv2026

MolSpecFlow: Mass-Constrained Hybrid Flow Matching for Joint Molecular-Spectral Analysis

Identifying the "dark matter" of the chemical universe requires bridging the fundamental gap between molecules and mass spectra. Existing approaches struggle to accurately map between these distinct data formats, often yielding results that are either chemically implausible or inconsistent with the physical evidence. We introduce MolSpecFlow, a unified foundation model pretrained on 100 million molecules and 42 million spectra, which leverages a hybrid flow matching framework to orchestrate optimal transport paths for spectral peaks and discrete probability flows for molecular tokens. Furthermore, to guarantee the physicochemical validity of the generated structures, we incorporate explicit rule-based constraints into the generative process, utilizing a token-level mass control mechanism that strictly enforces alignment with the precursor molecular weight. MolSpecFlow establishes a new state-of-the-art on the MassSpecGym benchmark, surpassing diffusion and retrieval baselines in de novo generation (Top-1 accuracy +35%), spectral simulation, and molecular retrieval, demonstrating the power of physics-grounded unified modeling.

bioinformatics↗

bioRxiv2026

Decoding the MYC locus reveals a druggable ultraconserved RNA element

The human genome is dominated by noncoding sequences, most of which are poorly conserved across species. How genetic information is distributed between coding and noncoding regions remains a fundamental unresolved question. Using CRISPR saturation mutagenesis at base-pair resolution, we mapped the functional fitness landscape of the 10-kb human MYC locus with a near-PAMless, high-fidelity SpRY-Cas9. This unbiased interrogation revealed that the majority (67%) of functionally essential base-pairs in this locus are noncoding. Paradoxically, the phenotypic impact of noncoding sequences correlates inversely with evolutionary conservation, driven in part by rapidly diverging cis-regulatory DNA elements that remain functionally constrained in humans. Within this landscape, we identified an ultraconserved RNA element in the 3 untranslated region (UTR) that is indispensable for MYC-dependent cancer cells. Remarkably, steric-blocking antisense oligos targeting this RNA element selectively eliminate MYC-addicted cancer cells by suppressing MYC function without reducing MYC abundance. Mechanistically, this 3' UTR element promotes perinuclear localization of MYC mRNA and efficient nuclear import of the short-lived MYC protein, enabling its function as a nuclear transcription factor. Together, these findings highlight noncoding sequences as major carriers of functional genetic information, provide a comprehensive fitness map of the MYC locus, and uncover a therapeutically actionable RNA element that disables MYC-driven cancer.

genomics↗

bioRxiv2026

caRBP-Pred: Deep Learning-based Prediction of Chromatin-Associated RNA-Binding Proteins Using Short Peptide Sequences

RNA-binding proteins (RBPs) are pivotal in cellular processes ranging from RNA metabolism to 3D genome organization. A distinct subset, chromatin-associated RBPs (caRBPs), binds directly to chromatin to function as transcriptional regulators. However, identifying caRBPs via traditional methods like Chromatin Immunoprecipitation Sequencing (ChIP-seq) and Mass Spectrometry (MS) is labor-intensive and costly. While computational tools for DNA- and RNA-binding protein (DRBP) prediction exist, they often rely on outdated Gene Ontology annotations and fail to capture the unique characteristics of chromatin association. Here, we introduce caRBP-Pred, a novel deep learning approach combining Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory networks (BiLSTM). Unlike previous methods utilizing full-length sequences, our model is trained on chromatin-contact peptides derived from mouse embryonic stem cells (mESCs). caRBP-Pred achieves a superior Area Under the Curve (AUC) of 0.81 using peptide sequence information alone, significantly outperforming existing DRBP predictors which exhibit low recall for caRBPs. We further predicted 52 potential caRBPs in mice. Notably, validation against human homologs confirmed that our model accurately predicts candidates with experimentally verified chromatin-binding capabilities. Collectively, caRBP-Pred is the first tool specifically designed to predict caRBPs based on chromatin-contact peptides, offering a valuable resource for investigating regulatory roles of caRBPs on transcription.

bioinformatics↗

bioRxiv2025

A Generative Foundation Model for Cryo-EM Densities

Single-particle cryo-electron microscopy (cryo-EM) enables structure determination of macromolecular complexes, yet reconstructions are often degraded by noise, anisotropic sampling and reconstruction artifacts. Correcting these effects is an ill-posed inverse problem that requires strong prior knowledge in addition to experimental data. Here we introduce cryoFM, a generative foundation model for cryo-EM densities that unifies data-driven structural priors with dataset-specific constraints within a Bayesian inference framework. CryoFM is trained in an unsupervised manner on thousands of high-quality cryo-EM maps using flow matching, learning a generalizable prior over macromolecular density distributions. Combined with explicit likelihood models describing experimental degradations, cryoFM enables flow posterior sampling, an inference-only procedure that performs denoising and restoration and refinement while remaining explicitly constrained by dataset-derived statistics. We show that this framework improves density reconstruction and refinement across diverse experimental settings, including preferred-orientation datasets and cases with strong spatial heterogeneity in signal-to-noise ratio, without introducing hallucinated features. In addition, cryoFM can be fine-tuned into conditional generative models for density post-processing, yielding maps with improved interpretability and fewer artifacts compared to existing supervised approaches. Together, cryoFM establishes generative foundation models as a principled and controllable framework for cryo-EM density reconstruction and modification.

biophysics↗