Biology subjects

Song, L.

Publications and source records attributed to Song, L..

At least 37 records · Page 2Linked to original sources

bioRxiv2026

Retroelement Hypomethylation Links Hypoxia Signaling, Immune Phenotypes, and Survival in Clear Cell Renal Cell Carcinoma

BackgroundRetrotransposable elements (RE) comprise approximately 45% of the human genome and are typically repressed by DNA methylation to preserve genomic integrity. In cancer, global DNA hypomethylation can lead to RE derepression, resulting in genomic instability and activation of innate immune pathways through viral mimicry. While individual RE classes have been examined in clear cell renal cell carcinoma (ccRCC), the integrated epigenetic landscape of multiple RE families and their clinical relevance remain incompletely characterized. MethodsWe performed a genome-wide prediction of DNA methylation across three major RE classes (Alu, LINE-1, and LTR elements) using a validated computational framework applied to Illumina methylation array data from two independent ccRCC tumor cohorts. Integrated unsupervised clustering of RE methylation profiles was used to define the epigenetic subtypes. Associations with clinicopathologic variables, tumor immune microenvironment composition (DNA Methylation-derived), hypoxia signaling, innate immune activation, and overall survival were evaluated. Prognostic relevance was assessed using multivariable Cox regression models adjusting for age, sex, AJCC stage or AUA risk group, and immune and angiogenic tumor microenvironment features. Key findings were then externally validated in CPTAC-ccRCC and independently replicated in an institutional Dartmouth Cancer Center (DCC) cohort with matched methylation and RNA-sequencing data. ResultsIntegrated clustering identified three reproducible RE methylation subtypes, Repressed, Transient, and Active. In the discovery cohort, the Active subtype showed significantly worse overall survival than the Repressed subtype, with a graded survival pattern across RE methylation states that persisted after multivariable adjustment. RE hypomethylation was associated with reduced EPAS1 (HIF2A) expression, increased immune infiltration, elevated PD-1 expression, and heightened cGAS-STING and interferon signaling, consistent with an immune-inflamed yet immunosuppressed tumor state. In the external CPTAC validation cohort, RE methylation subtypes recapitulated key molecular features and showed supportive survival trends. In the independent DCC replication cohort, an Active RE state was again associated with poorer survival, lower EPAS1 expression, increased PD-1 expression, greater CD8 T-cell and Treg infiltration, and elevated T-cell exhaustion signatures, supporting the reproducibility of the prognostic and immune-exhausted phenotype across cohorts. ConclusionsWe identified RE methylation subtypes with distinct molecular, immunologic, and prognostic features in ccRCC. External validation in CPTAC and independent replication in DCC support the robustness of this RE methylation framework across large-scale and institutional cohorts. These findings highlight the prognostic potential of RE methylation profiles and support their integration into molecular classification strategies to improve risk stratification in ccRCC.

cancer biology↗

bioRxiv2026

Large Language Model Agent Enables Autonomous Machine Learning Model Building for Biomedicine

Machine learning accelerates biomedical discovery, but creating effective predictive models requires specialized human expertise and demanding manual effort. Researchers must iteratively design pipelines, select architectures, and debug code. This challenge is particularly severe in biomedicine because of the heterogeneous datasets, sparse annotations, and complex evaluation protocols that are common in the domain. We present AIDO.Builder, an agentic artificial intelligence system that fully automates the entire life-cycle of biomedical model development. Provided only with a natural language task description and a target metric, AIDO.Builder autonomously constructs executable training and evaluation pipelines. The system selects suitable modeling strategies, executes experiments, and uses automated feedback-loop to iteratively revise its own code, configurations, and training procedures. It flexibly adapts to new tasks by training specialized models de novo or by using pretrained foundation models to build predictive models through task-appropriate adaptation. We show that across diverse biomedical benchmarks, AIDO. Builder produces highly competitive solutions against human alternatives, while eliminating the manual iteration previously required for robust model development. By automating the translation of raw data into reliable AI models, AIDO.Builder demonstrates how AI itself can be used to accelerate AI for biomedical research.

bioinformatics↗

bioRxiv2026

Harnessing AI to Build Virtual Cells

A virtual cell is a world model of a cell: a computational system that predicts, simulates and programs cellular processes across modalities and scales. An important path toward this goal is to model how genetic and chemical perturbations give rise to transcriptional responses, a core capability for disease understanding and drug discovery. However, current approaches remain expert-intensive, relying on iterative manual model design, training and debugging over months. Here we present VCHarness, an autonomous AI system that constructs perturbation-response models by combining an AI coding agent with multimodal biological foundation models. The system explores large spaces of architectures and training pipelines with minimal human intervention, iteratively generating, evaluating and refining candidate models. Across multiple perturbation-response benchmarks, VCHarness identifies architectures that outperform expert-designed approaches while reducing development time from months to days. It further uncovers non-obvious architectural patterns associated with improved performance, indicating that automated search can extend beyond conventional design strategies. These results suggest a shift from manually engineered models toward autonomous systems for constructing components of virtual cell world models, enabling scalable and data-driven exploration of cellular systems.

bioinformatics↗

bioRxiv2026

Detecting context-dependent selection on cancer driver genes with DiffDriver

Positive selection on somatic mutations is the driving force for cancer progression. Growing evidence shows that the emergence of a driver mutation in a tumor sample depends on individual-specific factors, for example environmental exposures or the individuals germline genetic background. We term these individual-level factors as the "contexts" of a tumor. Our hypothesis is that mutations in a driver gene can bring different growth advantages in different contexts, resulting in "differential selection" on these genes in varying contexts. Identifying which contexts modulate selection strength provides critical insights into the selection forces driving tumorigenesis. However, due to the sparsity of somatic mutations and heterogeneous background mutational process across positions and individuals, identification of differential selection has limited power with current statistical tools and is prone to false positives. To address this, we developed a powerful statistical method, DiffDriver, that identifies associations between "contexts" and selection strength on a driver gene across individuals. DiffDriver accounts for variations of mutation rates across bases and individuals, while taking advantage of functional information of sequences to improve the power. Through simulations, we show DiffDriver reduces false positives and boosts power compared to current methods. Our results highlight that multiple individual-level factors create significant heterogeneity in the strength of selection acting on driver genes and 33% of driver genes showed differential selection in at least one of the contexts studied, including tumor clinical traits and tumor immune microenvironment subtypes. These results provided new insights into the context-dependent forces driving cancer evolution.

genomics↗

bioRxiv2026

Comprehensive drug efficacy data for mucinous ovarian carcinoma using a novel and extensive biobank of patient-derived organoid models

Mucinous Ovarian Carcinoma (MOC) is a rare ovarian cancer histological subtype with distinct pathology, genomics and clinical outcomes compared to other epithelial ovarian cancers. Accordingly, there is little evidence to guide clinical care, particularly in the use of systemic therapies, and the field has lacked informative and diverse pre-clinical models. We developed MOC-specific methods for generating tumour organoids with a success rate of 70% for long-term cultured lines (n=19). Organoid lines were developed from localised, advanced and recurrent tumours, including from biopsy tissue, and represent diverse genomic features not previously captured by existing cell lines. The organoid lines were highly similar to the tumours of origin for genomic and immunohistochemical markers. Screening using a panel of 11 chemotherapy agents highlighted resistance to standard-of-care agents such as carboplatin. Gastrointestinal cancer chemotherapy agents and their combination regimens lacked activity. Paclitaxel was often highly potent at low doses but failed to kill all cells. However, less frequently used drugs such as gemcitabine, topotecan and doxorubicin inhibited many of the lines more effectively than paclitaxel. Available, but non-standard of care, chemotherapy agents should be considered for clinical management of MOC. This is the largest (by [~]10 fold) cohort of fully characterised patient-derived MOC organoid lines described and the first with extensive drug screening data affording an opportunity for drug discovery and screening for personalised treatment.

cancer biology↗

bioRxiv2026

GenBio-PathFM: A State-of-the-Art Foundation Model for Histopathology

Recent advancements in histopathology foundation models (FMs) have largely been driven by scaling the training data, often utilizing massive proprietary datasets. However, the long-tailed distribution of morphological features in whole-slide images (WSIs) makes simple scaling inefficient, as common morphologies dominate the learning signal. We introduce GenBio-PathFM, a 1.1B-parameter FM that achieves state-of-the-art performance on public benchmarks while using a fraction of the training data required by current leading models. The efficiency of GenBio-PathFM is underpinned by two primary innovations: an automated data curation pipeline that prioritizes morphological diversity and a novel dual-stage learning strategy which we term JEDI (JEPA + DINO). Across the THUNDER, HEST, and PathoROB benchmarks, GenBio-PathFM demonstrates state-of-the-art accuracy and robustness. GenBio-PathFM is the strongest open-weight model to date and the only state-of-the-art model trained exclusively on public data.

bioinformatics↗

bioRxiv2026

Foundation Models Improve Perturbation Response Prediction

Predicting cellular responses to genetic or chemical perturbations has been a long-standing goal in biology. Recent applications of foundation models to this task have yielded contradictory results regarding their superiority over simple baselines. We conducted an extensive analysis of over 600 different models across various prediction tasks and evaluation metrics, demonstrating that while some foundation models fail to outperform simple baselines, others significantly improve predictions for both genetic and chemical perturbations. Furthermore, we developed and evaluated methods for integrating multiple foundation models for perturbation prediction. Our results show that with sufficient data, these models approach fundamental performance limits, confirming that foundation models can improve cellular response simulations. Code and Data: https://github.com/genbio-ai/foundation-models-perturbation

bioinformatics↗

bioRxiv2026

Improving Growth Predictions in Aquaculture through an Improved Bioenergetics Model Incorporating Feed Composition and Nutrient Digestibility for Largemouth Bass (Micropterus salmoides)

Bioenergetics models serve as mechanistic tools to predict growth by linking energy intake, metabolic expenditure, and nutrient partitioning. However, traditional models rely primarily on gross energy (GE) intake, thereby oversimplifying the effects of feed composition and nutrient availability on fish growth. This work therefore proposed a refined bioenergetics model incorporating nutrient-specific digestibility coefficients (ADCs) and feed composition and tested using a compiled dataset (n=235; 165 for calibration and 70 for independent validation) and a field experimental dataset of largemouth bass (Micropterus salmoides). We first optimized parameters of a gross energy intake-based bioenergetics model, increasing R2 from 0.62 to 0.96 and thereby providing a calibrated foundation for subsequent refined model. The refined model demonstrated superior predictive performance on the compiled dataset (R2 = 0.97) with RMSE = 19.86 g and MAE = 10.31 g), representing reductions of 4.13% in RMSE and 19.98% in MAE and a 1.03% increase in R2 compared with the optimized GE-based model. In the field experiment, the refined model achieved high predictive accuracy (R2 = 0.98 and 0.97), whereas the optimized GE-based model showed poor performance (R2 = 0.33 and 0.06 respectively). This study is, to our knowledge, the first bioenergetics framework for largemouth bass that decomposes feed composition and nutrient-specific ADCs to compute macronutrient-resolved digestible energy, enabling formulation-aware growth prediction and nutrient-oriented optimization.

bioengineering↗

bioRxiv2026

Phosphorylation of Cyclophilin-D is Not Required for Regulation of The Mitochondrial Permeability Transition Pore by GSK3β

Genetic inhibition of cyclophilin D (CypD) delays the opening of the mitochondrial permeability transition pore (MPTP) and therefore reduces necrotic cell death. Elucidation of factors that impact CypD activity is therefore key to understanding the regulation of MPTP opening. Glycogen synthase kinase-3{beta} (GSK3{beta}) is a serine/threonine kinase that has been shown to modulate MPTP and cell death, potentially through phosphorylation of CypD. Therefore, we hypothesized that the mitochondrial fraction of GSK3{beta} directly phosphorylates CypD and promotes opening of MPTP. Overexpression of full length GSK3{beta} in mouse embryonic fibroblasts sensitized the MPTP and exacerbated oxidative stress-induced necrosis. In contrast, genetic inhibition of GSK3{beta} protected against oxidant-induced cytotoxicity but did not affect the MPTP. Recombinant GSK3{beta} could directly bind to and phosphorylate recombinant CypD. Mass spectrometry revealed several putative GSK3{beta} phosphorylation sites on CypD. However, mutation of these sites did not affect the peptidyl prolyl isomerase activity of CypD and reconstitution of these phosphomutants in CypD-deficient cells increased MPTP sensitivity and oxidative-induced cell death to the same extent as wild-type CypD. Further, targeted overexpression of either wild-type or kinase-inactive GSK3{beta} in the mitochondrial matrix did not impact MPTP or cell death. Moreover, while proteinase-K digestion of cardiac mitochondria showed a significant amount of GSK3{beta} in the mitochondria, it was not localized to the matrix. Finally, overexpression of GSK3{beta} was still able to increase MPTP sensitivity and oxidative stress-induced death in CypD-null cells. Taken together, these data indicate that, while GSK3{beta} can modulate MPTP, this appears to be independent of GSK3{beta}s interaction with, or phosphorylation of CypD.

cell biology↗

bioRxiv2025

Multimodal AI for Single cfDNA Profiling and Cancer Screening

Cell-free DNA (cfDNA) serves as a non-invasive biomarker for cancer detection, but conventional methods face challenges due to the ultra-low abundance of tumor-derived cfDNA (ctDNA) among normal cfDNA. Though nucleosome-bound cfDNA harbors rich epigenomic features that could enable ctDNA identification by single-molecule multi-omics cross-validation, this remains unexplored due to methodological limits. Here, we developed a cfDNA sequencing approach integrating methylation, fragmentomics, and histone modifications at the single-molecule level; together with gene semantics and epigenomic annotations, these modalities were vectorized and fused to represent each cfDNA molecule. We trained a Transformer-based model (cfAI) to profile and evaluate ctDNA likelihood at molecule, gene, and sample levels. cfAI achieved [~]10-fold enrichment of cancer-derived signals over noise and reached 72.6% sensitivity at 93.1% specificity for multi-cancer detection. Our study establishes an innovative framework that overcomes the inherent signal-to-noise limitations of conventional assays and reveals biological features at molecular resolution for cancer detection.

bioinformatics↗

bioRxiv2025

TLR3 Activation of Microglia-Containing Cerebral Organoid Induces Antiviral Factors against HIV-1 Infection

Human induced pluripotent stem cell (iPSC)-derived cerebral organoids have been increasingly used as a brain model for studying various neurological disorders and neurotropic virus infections, including HIV-1. However, it is unclear whether iPSC-derived cerebral organoids possess functional innate antiviral immunity against HIV-1. In this study, we examined Toll-Like Receptor 3 (TLR3) activation and its role in the induction of IFNs and IFN-stimulated genes (ISGs) against HIV-1 in human iPSC-derived microglia containing cerebral organoids (MCOs). We observed that MCOs possess functional TLR3, which could be effectively activated by poly I:C. TLR3 activation of MCOs resulted in HIV-1 inhibition and induction of IFN-{beta}/IFN-{lambda}, antiviral ISGs (MX1, MX2, GBP5, SAMHD1, Viperin, and ISG56), and CC chemokines (MIP-1, MIP-1{beta}, and RANTES), the ligands for HIV entry coreceptor CCR5. This TLR3 activation-mediated anti-HIV-1 effects could be blocked by a specific TLR3 inhibitor. These findings indicate that human iPSC-derived MCOs are a suitable model for investigation of brain immunity and HIV-1 infection.

immunology↗

bioRxiv2025

A Globally Representative Immunopeptidomics Approach to Identify Population-Wide Vaccine Candidates

Vaccine development has historically relied on preclinical testing using one or two cell lines, typically of Caucasian descent, contributing to vaccines that may fail to provide adequate protection across genetically diverse populations. A major factor in these failures is the lack of consideration for human leukocyte antigen (HLA) diversity, with over 30,000 HLA alleles worldwide exhibiting distinct frequencies across ethnic groups and geographic regions. Here, we present a globally representative immunopeptidomics approach that addresses HLA diversity at the earliest stages of vaccine antigen discovery. We established a panel of 30 cell lines from the 1000 Genomes Project that represent 75% of the worlds most common HLA alleles. Using a combined computational and experimental approach, we identified 18 common binding motifs (7 MHC class I and 11 MHC class II) shared across 24 cell lines from the panel. We developed an accessible bioinformatic tool that predicts globally presentable antigenic regions from any pathogen proteome by identifying peptides matching these common binding motifs. Validation with Salmonella enterica serovar Typhimurium and SARS-CoV-2 Spike protein demonstrated that our method successfully identifies known immunogenic regions, with experimentally detected epitopes showing substantial overlap with predicted regions of interest. Notably, only three cell lines were required to validate highly immunogenic S. enterica targets, including OmpC--a porin with known 100% protective efficacy--detected across all tested lines. Our approach enables researchers to rapidly screen pathogen proteomes using freely available bioinformatic tools, with optional experimental validation requiring minimal resources. By identifying antigens with broad population coverage before clinical trials begin, this method has the potential to increase vaccine success rates while ensuring equitable protection across diverse populations. The cell line panel, common binding motifs, and bioinformatic workflow are publicly available, providing an accessible pathway for developing vaccines that truly serve global populations.

immunology↗

bioRxiv2025

PertAdapt: Unlocking Single-Cell Foundation Models for Genetic Perturbation Prediction via Condition-Sensitive Adaptation

Single-cell foundation models (FMs) pretrained on massive unlabeled scRNA-seq data show strong potential in predicting transcriptional responses to unseen genetic perturbations. However, existing approaches insufficiently transfer pretrained knowledge and overlook the imbalance between perturbation-sensitive and insensitive genes, yielding only marginal improvements over nonpretrained baselines. To address these limitations, we introduce PertAdapt, a framework that unlocks FMs to accurately predict genetic perturbation effects via integrating a plug-in perturbation adapter and an adaptive loss. The adapter employs a gene-similarity-masked attention mechanism to jointly encode perturbation conditions and contextualized representations of unperturbed cells, enabling more effective knowledge transfer. To better capture differential expression patterns, the adaptive loss dynamically reweights perturbation-sensitive genes relative to global transcriptomic signals. Extensive experiments across seven perturbation datasets, including both single- and double-gene settings, demonstrate that PertAdapt consistently outperforms non-pretrained and FM baselines. Moreover, PertAdapt demonstrates strong capacity for modeling multiplexed gene interactions, generalizing in limited-data regimes, and maintaining robustness across backbone sizes. AvailabilityCode and data are available at https://github.com/BaiDing1234/PertAdapt.

bioinformatics↗

bioRxiv2025

Challenges and opportunities in detecting leaf water and carotenoid content across biomes from satellite multispectral indices

Climate change is causing vegetation stress across the globe, increasing the need for reliable indicators to monitor plant health. Leaf water and carotenoid content, and the chlorophyll/carotenoid ratio, are established proxies for environmental stress that can be detected by remote sensing. Here, we evaluated the sensitivity of 11 multispectral vegetation indices (VIs) designed to monitor these three stress-related leaf traits across a broad range of environmental and vegetation conditions. For this, we combined radiative transfer modeling with cross-biome field and satellite observations from Sentinel-2, Landsat 8, and MODIS from the National Ecological Observatory Network (NEON), spanning in most major terrestrial ecosystems. Our model-based analysis showed that VIs have a low to moderate sensitivity to their target traits, ranging from water indices with 66% of their variability explained by leaf water content, to carotenoid indices with 27% variability explained by leaf carotenoid content. Surprisingly, our field-based analyses revealed minimal to no sensitivity to leaf water and carotenoid content and chlorophyll/carotenoid ratio across all VIs. In contrast, we showed that leaf area index was the dominant driver of all studied VIs, accounting for 54-74 % of their variability in the field-based analysis. Lastly, we detected that VI[s] sensitivity to atmospheric conditions and field sampling issues contribute to their low performance in validating ground truth observations. These findings show that improvements in the VIs formulation and field sampling strategies are needed to increase the reliability of vegetation stress monitoring from multispectral satellites and support a generalized use of VIs across ecosystems. Highlights: 3-5 bullet points, 85 characters[bullet] Sensitivity of water and carotenoid multispectral indices was evaluated [bullet]Analysis based on cross-biome field data and radiative transfer models [bullet]Field data showed indices had minimal sensitivity to leaf water and carotenoid [bullet]Leaf area index explained most cross-biome variation in water and carotenoid indices [bullet]We propose strategies to improve stress-related index formulation and validation

ecology↗

bioRxiv2025

Web engine for tumor pathology image retrievals on massive scales

Hematoxylin and Eosin staining (H&E) is widely used in clinical practice, but efficient and versatile image retrieval tools are lacking. We developed the H&E Retrieval Engine (HERE, https://hereapp.ccr.cancer.gov) to analyze patient cases based on image similarities to database records. Using H&E image regions as input, HERE searches 21.2 terabytes of whole-slide images from multiple tumor histopathology cohorts through a 12.1-gigabyte memory index, and returns top images containing regions similar to the query. HERE scans high-resolution images in the database using accurate artificial intelligence encoding and ultra-efficient hierarchical skip indexing. HERE demonstrated performance superior to existing image retrieval tools based on blinded pathologist scoring using benchmark queries that represent key image features of human tumors. By pairing spatial transcriptomics with H&E images, HERE also enables retrieving image features from gene transcriptomics input and identifies molecular pathways associated with tumor histologies.

bioinformatics↗

bioRxiv2025

Comprehensive profiling of transcription factors for reprogramming human astrocytes to neuronal cells through endogenous CRISPR-based gene activation

Neuronal loss is a hallmark of neurodegeneration and brain injury. Direct reprogramming of astrocytes into neurons has emerged as a promising approach to restore lost neurons. Comprehensive mapping and characterization of candidate astrocyte-to-neuron reprogramming factors is an essential step to realizing the potential of this strategy. Here, we established a CRISPR activation (CRISPRa)-based approach for neuronal reprogramming of primary human astrocytes. We conducted high-throughput CRISPRa screens of all human genes encoding transcription factors (TFs) to identify novel and efficient reprogramming factors. scRNA-seq characterization of top hits revealed that single TFs reprogram primary human astrocytes into multiple neuronal subtypes with distinct cell type-specific gene signatures. We demonstrate that INSM1 reprograms astrocytes to a glutamatergic neuron-like state and has broad neurogenic activity across different cell types and across human and mouse contexts. Finally, we conduct paired CRISPRa screens to identify cofactors that cooperate with INSM1 to enhance neuronal reprogramming and subtype specification, and elucidate genomic mechanisms of interaction and downstream regulators.

cell biology↗

bioRxiv2025

SlATG8f enhances tomato thermotolerance and fruit quality via autophagy and HS pathways

Plant responses to high-temperature stress involve dynamic and complex changes in physiology, biochemistry, and gene expression. Autophagy, a highly conserved intracellular degradation system, facilitates the removal and recycling of damaged cytoplasmic components such as proteins and organelles. It plays an essential role in plant growth, development, and adaptation to various stresses. Although the functions of many autophagy-related genes under high-temperature stress have been characterized, the role of SlATG8f, a key member of the SlATG8 family in tomato, remains poorly understood. In this study, we generated SlATG8f-Overexpressing tomato lines using the recombinant vector pBWA(V)HS. Using quantitative reverse transcription polymerase chain reaction (qRT-PCR), we analyzed physiological indices and expression levels of ATG8 family members and heat shock protein-related genes in fruits of wild-type (WT) and SlATG8f-overexpressing plants across four developmental stages under high-temperature stress. Our results indicate that overexpression of SlATG8f upregulates the expression of autophagy-related and heat shock protein genes, promotes early fruit ripening, and improves fruit quality under high-temperature stress. These findings reveal a regulatory role for SlATG8f in maintaining tomato fruit quality under heat stress and provide a theoretical foundation for tomato variety improvement.

plant biology↗

bioRxiv2025

AtSDR4L and its paralog DIG2 repress somatic embryogenesis during post-embryonic development in Arabidopsis

Somatic embryogenesis, a developmental process advantageous as an alternative means for plant propagation due to the lack of maternal tissues and the ease of in vivo to in vitro knowledge transfer of embryogenesis, has been applied in studies on the regulatory mechanisms governing the developmental plasticity in plants. Core and accessory components of the polycomb repressive complexes (PRCs), the epigenetic regulatory machinery, are known to repress the transcription of genes that positively regulate the somatic embryogenesis in seedlings. Recent studies revealed that Arabidopsis thaliana SEED DORMANCY 4-LIKE (AtSDR4L), an interactor of a PRC accessory, represses the embryo developmental program for a successful switch from seed to seedling. However, little is known about the regulation of cell differentiation by AtSDR4L in seedlings. Here, we show that AtSDR4L, its paralog Dynamic Influencer of Gene expression 2 (DIG2), and PRC components share similar transcriptional regulatory dynamics, particularly on somatic embryogenesis-related genes. We further demonstrate that mutations in AtSDR4L and DIG2 lead to the formation of somatic embryo-like structures and a delay in cellular differentiation. Thus, transcriptional changes reflected by morphological defects associated with somatic embryogenesis in Atsdr4l dig2 mutants suggest that AtSDR4L and DIG2 prevent the dedifferentiation of cells for proper seedling growth through the transcriptional control of cell identity.

plant biology↗