AI-driven framework modeling perturbation in brain organoids reveals candidate genes for autism

Interpretable gene networks from single-cell foundation models reveal conserved neurogenic dysfunction in Parkinson’s disease

Interpreting large-scale single-cell transcriptomic data remains a major challenge for understanding disease mechanisms. Recent single-cell foundation models learn rich representations of gene relationships across millions of cells, yet methods for translating these embeddings into biologically interpretable gene networks remain limited. Here we present scGENet, a computational framework that constructs context-specific gene interaction networks from foundation model–derived gene embeddings. By fine-tuning pretrained models on transcriptomic data from human midbrain organoids, scGENet generates transcriptome-scale gene modules that capture biologically meaningful cellular programs. Benchmarking across multiple foundation models demonstrates that networks derived from a fine-tuned scGPT brain model show the highest concordance with curated neuronal pathways, Parkinson’s disease (PD) genetic risk loci, and independent patient-derived transcriptional signatures. Applying this framework to human iPSC-derived PD midbrain organoids reveals transcriptional modules associated with neuronal differentiation, synaptic signaling, and cell-cycle regulation. Single-nucleus RNA sequencing further links these programs to altered cellular composition, including reduced dopaminergic neurons, expansion of radial glia–like progenitors, and a dopaminergic neuron subtype expressing SNCA and VGLUT2. Integration with independent human substantia nigra datasets identifies a conserved neurogenic program disrupted across genetic and idiopathic PD. Together, these results establish a generalizable strategy for extracting interpretable gene networks from single-cell foundation models, enabling systematic discovery of disease-relevant molecular programs across diverse tissues and datasets.

AI-Guided Systems Neurogenomics in Neurodevelopmental Disorders

Despite substantial advances in genomic testing, many individuals with neurodevelopmental disorders remain without a molecular diagnosis, while others receive a genetic diagnosis that does not fully explain phenotypic variability, developmental trajectory or tissue-specific consequences. Artificial intelligence (AI)-assisted methods are increasingly used for phenotyping, variant prioritisation, splice prediction, protein modelling, DNA methylation episignature classification and multi-omic analysis. However, these approaches differ substantially in evidentiary status and are often applied as separate prediction tasks rather than as components of an explicit mechanistic model. In this targeted narrative review, focused primarily on rare and genetically enriched neurodevelopmental disorders, we examine how AI-assisted methods may contribute to systems-level interpretation while remaining anchored to established molecular diagnosis and variant-classification frameworks. We propose a hypothesis-generating load-capacity framework comprising regulatory load, network capacity, developmental buffering and regulatory network instability. These are treated as operationalisable but currently unvalidated constructs. Regulatory instability is distinguished from stable disease-associated dysregulation, and threshold-like behaviour is presented as an empirical possibility rather than an assumed property of neurodevelopmental disease. We formulate five falsifiable predictions, consider how genomic, transcriptomic, epigenomic, single-cell, spatial, imaging, neurophysiological and longitudinal phenotypic evidence can provide complementary mechanistic constraints, and outline an auditable workflow following nondiagnostic genomic testing. We distinguish clinically implemented approaches from translational, emerging and conceptual applications, and emphasise calibration, evidence traceability, domain validity, prospective validation and appropriate abstention. Finally, we describe the Instability Twin as a prospective architecture composed of independently testable patient-specific sub-models rather than an existing clinical platform. The central proposition is that systems neurogenomics should be evaluated by whether mechanistically constrained integration provides reproducible information beyond established gene-level and simpler multimodal approaches.

Systematic CRISPRi perturbation of 1,408 autism risk genes maps multilevel transcriptional convergence in human cortical neurons

Genetic heterogeneity in autism spectrum disorder (ASD) complicates the identification of shared molecular pathways amenable to therapeutic intervention. Here, we perform a CRISPRi Perturb-seq screen targeting 1,408 ASD risk genes in human embryonic stem cell-derived immature cortical neurons and profile the resulting transcriptomic effects by single-cell RNA sequencing. We identify 215 ASD risk genes whose repression induces significant global transcriptomic dysregulation. Leveraging this functional atlas, we characterize candidate genes based on transcriptional similarity to established ASD hub genes. We identify gene programs recurrently dysregulated across perturbations, anchored by processes governing microtubule dynamics, neuron differentiation, cell migration, and transmembrane transport. We further identify CHAMP1 as a previously unrecognized regulator of Wnt signaling and define specific ASD risk genes that modulate the rate of cortical neuron differentiation. Analysis of differentially expressed genes reveals both convergent and perturbation-specific downstream transcriptional responses. Together, these findings provide a multilevel map of transcriptomic convergence in ASD and establish a framework for identifying both pathway-level and genotype-specific therapeutic strategies.

Developmental and cellular vulnerabilities underlie genetic architecture of schizophrenia.

Molecules and Cells

Multi-omic data integration improves the resolution of the molecular etiology of autism in a mouse model.

RNA-sequencing, 3-dimensional protein-centric chromatin conformation, and whole genome DNA methylation sequencing approaches are used to investigate hippocampal tissue from an ASD mouse model to determine if multi-omic data integration improves the resolution of key molecular pathways contributing to the complex ASD phenotype.