The problem veloAgent tackles is well documented. Existing tools such as scVelo, cellDancer, DeepVelo, veloVI, VeloVAE, UniTVelo, and SIRV each improve some aspect of velocity estimation, but most assume transcriptional kinetics that are uniform across cells, an oversimplification that erases the heterogeneity driving development and disease. More sophisticated models that estimate gene- and cell-specific rates, like cellDancer, become computationally prohibitive on atlas-scale datasets. And nearly all methods rely on embeddings of transcriptional similarity alone, discarding the tissue microenvironment. The result is velocity fields that contradict known tissue structure, misdirect flows from tumor cells toward healthy neighbors, or push neurons toward neuroblasts they can never become.

veloAgent is a hybrid of three components. First, a variational autoencoder compresses noisy spliced and unspliced count matrices from each cell into a shared latent representation, denoising the data while preserving the features that mark dynamic cellular states. Second, a deep neural network constrained by protein–protein interactions from the STRING database takes that latent vector and infers transcription, splicing, and degradation rates—α, β, and γ—for every gene in every cell. The biologically informed sparsity matters: each gene’s output depends only on its annotated interactors, grounding the learned kinetics in known regulatory relationships. These parameters feed ordinary differential equations that yield initial velocity estimates, which are then refined by aligning each cell’s vector with those of its transcriptionally similar neighbors.

The third component is the distinctive one. An agent-based model treats every cell as an autonomous agent sitting at its true position in the tissue, and updates its velocity using local interaction rules. Three factors shape each update: expression similarity between neighbors, local cell density, and inverse spatial distance, with closer and more transcriptionally alike cells exerting stronger influence. The result is a velocity field that respects tissue architecture rather than floating free of it. Crucially, the authors emphasize that the projected arrows represent inferred transcriptional state transitions, not physical cell migration—a distinction that keeps the spatial maps honest while still revealing how tissue organization constrains and guides cellular trajectories.

The benchmarking is extensive. The team applied veloAgent to four spatial transcriptomics datasets spanning different platforms and resolutions: a Stereo-seq mouse brain, a human breast cancer Visium sample, a developing chicken heart, and a HybISS mouse brain. In the Stereo-seq brain, competing methods produced unclear or contradictory velocities around fiber-tract cells, while veloAgent captured coherent transitions from hippocampal, thalamic sensory-motor, and polymodal association regions consistent with known neuronal projections. In breast cancer, SIRV incorrectly directed flow from tumor cells toward normal cells; veloAgent corrected the misprediction. In the chicken heart, it fixed reversed trajectories at the ventricle–atrium boundary, and in the HybISS brain it correctly modeled neurons as terminal states rather than steering them toward neuroblasts.

Quantitatively, veloAgent was evaluated against seven leading methods on three metrics: cross-boundary direction correctness, fate probability, and velocity confidence. It achieved the highest or near-highest scores across all datasets. For cross-boundary direction, it outperformed all competitors by an average of 92.03 percent on the Stereo-seq brain, 92.02 percent on breast cancer, 55.52 percent on the chicken heart, and 12.15 percent on the HybISS brain. Fate probability gains averaged between 33 and 66 percent, and velocity confidence remained near-perfect, at 0.998 or above, across every dataset. Marker-gene analyses reinforced the picture: Mbp velocity stayed confined to myelinating fiber tracts, ERBB2 velocity localized to tumor regions rather than macrophages, NPPA velocity remained restricted to the atria, and Nrg1 velocity marked neurons specifically.

The improved velocities also sharpen downstream biology. Feeding veloAgent’s output into CellRank, the researchers recovered lineage-specific driver genes ordered correctly along pseudotime. In the developing mouse brain, early En1 expression aligned with progenitor states, Dlk1 peaked mid-trajectory alongside pro-neural factors, and late genes such as Lhx5 and Klhl14 marked terminal neuronal populations, with Pax8 emerging as a key mid-hindbrain regulator. In the chicken heart, transient valve progenitors enriched for DICER1 were distinguished from terminal valve cells expressing extracellular matrix genes like COL1A1 and SPARC. Gene Ontology enrichment confirmed the programs: circulatory-system development and mesenchymal development in the heart, neuron differentiation and nervous-system development in the brain. SCENIC analysis further recovered known transcription factors, including Foxj1, Lmx1a, Msx1, and Msx2 in choroid-plexus patterning, Nfia and Tcf12 in fiber-tract formation, and Foxf1 and Gata4 in endoderm and cardiac lineages.

The most provocative feature, however, is the in silico perturbation module—something no prior velocity framework offers. Because veloAgent explicitly parameterizes α, β, and γ for each gene, researchers can set a gene’s transcription rate to zero after training, without retraining, and recompute the entire velocity field. The change in cross-boundary directionality then quantifies that gene’s causal influence on fate progression. Applied to the mouse brain, silencing top-ranked genes reversed differentiation toward the fiber-tract fate, and the perturbed genes were enriched for neurogenesis, axonogenesis, and synaptic organization. Applied to breast cancer, the top hits included CARD14 and ERBB2, both established oncogenes whose virtual silencing redirected velocity vectors away from malignant clusters. Kaplan–Meier analyses showed that patients with high expression of these genes had significantly poorer overall survival, and ERBB2 is already the target of FDA-approved drugs including trastuzumab and trastuzumab deruxtecan—a striking validation of the computational predictions.

Scalability is the other headline advance. veloAgent’s architecture scales with the number of genes, which is essentially fixed per organism, rather than with cell count, which grows exponentially with sequencing throughput. When the team duplicated the HybISS dataset from 50,000 up to one million cells, cellDancer’s memory usage climbed near-cubically to roughly 120 gigabytes, while veloAgent required only about 37 gigabytes with near-constant per-cell runtime. The agent-based module is also fully modular: applied post hoc to scVelo’s velocity estimates, it lifted velocity confidence dramatically across all four datasets—for example from 0.0677 to 0.9943 on the Stereo-seq brain—without retraining the underlying model. Ablation experiments confirmed that removing any component, especially the spatial ABM, significantly degraded performance.

The authors are candid about limitations. Visium spots can capture transcripts from multiple cells, so inferred dynamics represent averaged states, though consistent results on higher-resolution platforms suggest the signals are robust. The perturbation framework currently handles single genes rather than combinatorial interventions, interaction databases like STRING remain incomplete, and the model operates at gene rather than exon level. It has also not yet been run on a true million-cell atlas. Even so, veloAgent marks a conceptual shift: RNA velocity becomes a spatially coherent, causally probeable model of tissue dynamics rather than an abstract trajectory plot. For developmental biologists mapping organogenesis, oncologists tracing tumor evolution, and regenerative medicine researchers seeking to steer stem cells toward desired fates, the ability to simulate a gene knockout in seconds and watch the cellular movie change direction could reshape how hypotheses are generated—and which experiments get run first.

Subject of Research: A spatially informed RNA velocity framework combining deep generative modeling and agent-based simulation to infer and perturb cell state transitions in tissues

Article Title: Dissecting and steering cell dynamics using spatially-informed RNA velocity with veloAgent

Article References: Raghavan, V., Yoon, B., Fonseca, G. J., Li, Y., & Ding, J. (2026). Dissecting and steering cell dynamics using spatially-informed RNA velocity with veloAgent. Molecular Systems Biology, 22(7), 1180-1200. https://doi.org/10.1038/s44320-026-00213-w

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00213-w

Keywords: RNA velocity, spatial transcriptomics, single-cell RNA sequencing, agent-based modeling, variational autoencoder, computational biology, cell fate, in silico perturbation, breast cancer, ERBB2, developmental biology, therapeutic target discovery

The core insight behind TFdisc is that wild-type single-cell transcriptomes already encode a surprising amount of information about how cells would respond if a regulatory gene were lost. The simulator works by first reconstructing gene regulatory networks from reference single-cell RNA sequencing data, mapping which transcription factors are likely to control which target genes based on co-expression patterns, regulatory motifs and expression dynamics across cell states. When a user specifies a transcription factor to perturb, TFdisc propagates the simulated disruption through this network, adjusting the expression of downstream targets and then recomputing the resulting cell states. The output is a predicted post-perturbation single-cell dataset: a virtual census of how each cell type would respond, which genes would shift their expression, and whether cells would drift toward or away from their normal identities and differentiation trajectories.

Validation was a central concern for the team, because a simulator is only useful if its predictions match reality. The researchers benchmarked TFdisc against multiple experimental perturbation datasets in which transcription factors had genuinely been knocked out or knocked down in the laboratory. Across these datasets, the model proved accurate in three demanding tasks: reconstructing the underlying gene regulatory networks, identifying the differentially expressed genes that change after perturbation, and predicting shifts in cell identity and lineage differentiation. The authors also compared TFdisc against existing perturbation-prediction tools, including approaches that require experimental knockout data for training, and found that their simulator, which can operate on wild-type data alone, performed competitively. Supplementary analyses covered imputation methods for sparse single-cell data, robustness of network construction, and sensitivity analyses designed to confirm that the findings were not artifacts of normalization or dataset size.

With the simulator validated, the team turned it toward one of the most stubborn problems in neuroscience: the genetic architecture of brain disorders. Conditions such as schizophrenia and autism spectrum disorder are influenced by large numbers of genetic risk variants, most of which individually contribute only a small amount of risk. Genome-wide association studies and sequencing efforts have catalogued hundreds of candidate risk genes, but converting those lists into mechanistic understanding has proven extraordinarily difficult. TFdisc offered a way to ask a question that would otherwise require decades of animal work: what happens to the developing and adult human brain, cell by cell, when each of these risk transcription factors is perturbed, alone or in combination?

The simulations revealed that different brain disorders are built on strikingly different genetic blueprints. When the researchers simulated the simultaneous perturbation of multiple schizophrenia risk transcription factors, the individual effects appeared scattered and heterogeneous, yet when combined they converged on a coherent set of shared molecular pathways. The team described this architecture as a jigsaw mechanism: each risk factor contributes one piece, and the disorder emerges only when many pieces are assembled together, with no single gene sufficient to produce the disease-relevant disruption. Autism spectrum disorder showed the opposite pattern. Perturbing its risk transcription factors produced overlapping, redundant effects in which individual factors each engaged a common core of pathways, an architecture the authors termed a monolithic mechanism. In this picture, many different genetic insults funnel into a similar biological outcome, which may help explain why autism can arise from such a wide variety of genetic lesions yet present as a recognizable clinical syndrome.

The simulator also traced how disease-relevant perturbation effects unfold across development. By applying TFdisc to single-cell reference data spanning prenatal development through adulthood, the researchers could identify the cell types and developmental stages in which simulated perturbations of risk factors produced their strongest molecular signatures. The supplementary materials document cell-type-stage-specific effects from prenatal development to adulthood, and independent validation of the predicted developmental regulatory programs using an external dataset confirmed that the simulator’s developmental predictions held up against real data. Shared pathways between schizophrenia and autism were detectable across developmental stages, but the timing and cellular context of peak pathway activity differed between the two disorders, reinforcing the conclusion that they follow distinct trajectories despite overlapping genetic risk.

Technically, the pipeline behind these findings involved several layers of quality control. The team benchmarked four imputation methods for handling the dropout and sparsity that plague single-cell RNA sequencing, evaluated gene regulatory network construction through robustness and sensitivity analyses, and assessed performance on simulated single-cell datasets where ground truth was known. Curated lists of transcription factors and risk genes associated with ten diseases were assembled from public databases and the literature, and pathway enrichment results were de-redundant to avoid inflating apparent biological signal with overlapping gene ontology terms. Clustering analyses of the risk transcription factors themselves helped organize the jigsaw and monolithic patterns, and phenotypic analyses connected the simulated perturbation clusters to known disease characteristics. This methodological scaffolding is important because perturbation prediction is a young field in which evaluation standards are still being established, and the Fudan team’s benchmarks provide a template for how future simulators should be assessed.

The implications for drug discovery and experimental design are considerable. A validated in silico perturbation simulator allows researchers to prioritize which transcription factors, cell types and developmental windows deserve the most intensive laboratory attention, dramatically narrowing a search space that would otherwise be prohibitive. For schizophrenia, the jigsaw architecture suggests that therapeutic strategies aimed at a single risk gene may be doomed to fail, and that interventions targeting the convergent downstream pathways, or combinations of factors, may be more promising. For autism, the monolithic architecture implies that a therapy correcting the shared core pathways could potentially benefit patients whose conditions arise from very different genetic causes. More broadly, the approach demonstrates that computational simulation can serve as a first-pass screen for perturbation biology, generating testable hypotheses about gene function at a scale unattainable in the wet laboratory.

The work, supported by funding from China’s National Key Research and Development Project, the National Science and Technology Innovation 2030 Major Program, the National Natural Science Foundation of China and the Shanghai Science and Technology Commission, arrives as single-cell biology and artificial intelligence converge on the problem of complex disease. The authors caution that their simulations are predictions, not proof, and that experimental validation of specific perturbation effects remains essential. Yet the study offers a compelling demonstration that the regulatory code written into ordinary single-cell data can be decoded to reveal how genetic risk is transformed into molecular dysfunction. As perturbation datasets accumulate and simulators like TFdisc are refined, the prospect of systematically mapping the causal paths from risk variant to altered cell state, and ultimately to disorder, moves from aspiration toward routine practice, promising a more mechanistic era for psychiatric genetics.

Subject of Research: An in silico transcription factor perturbation simulator that models gene regulatory responses in brain disorders using single-cell RNA sequencing data.

Article Title: An in silico transcription factor perturbation simulator uncovers diverse genetic architectures of brain disorders

Article References: Wang, H., Li, Q., & Zhu, Y. (2026). An in silico transcription factor perturbation simulator uncovers diverse genetic architectures of brain disorders. Genome Medicine. https://doi.org/10.1186/s13073-026-01752-5

Image Credits: AI Generated

DOI: 10.1186/s13073-026-01752-5

Keywords: transcription factors, single-cell RNA sequencing, gene regulatory networks, schizophrenia, autism spectrum disorder, brain disorders, in silico perturbation, TFdisc, genetic architecture, computational biology, Genome Medicine, Fudan University