A broader way to read genetic risk

Schizophrenia is not explained by a single faulty gene. It is a complex psychiatric disorder in which inherited variation, environmental influences and other factors interact. Many genetic variants associated with risk sit outside protein-coding DNA, making it difficult to determine which genes they affect and through what biological pathways.

That is the problem addressed by a study published in Nature Genetics in June 2026. The researchers used computational models to estimate how genetic variation may influence gene expression across interconnected networks in the brain. Their analysis identified 766 genes associated with schizophrenia, including 641 not highlighted in earlier transcriptome-wide studies.

The result is significant less because it settles the biology of schizophrenia than because it changes the search strategy. Rather than considering genetic effects largely one gene at a time and near each gene’s physical location in the genome, the work models more distant regulatory signals and the coordinated activity of genes that tend to be expressed together.

What the models add

Genome-wide association studies can find DNA variants statistically linked to a condition, but those signals often do not directly reveal the responsible gene or the mechanism involved. Transcriptome-wide association studies try to bridge that gap: they combine genetic association data with reference datasets that connect variants to gene expression.

Conventional approaches have focused predominantly on local, or cis, regulatory effects. These are variants located close to the gene whose expression they may influence. That is a practical starting point, but it leaves out trans effects, in which genetic variation can shape expression at genes elsewhere in the genome through wider regulatory systems.

The new study introduced two complementary frameworks, called INGENE and MODULE. Both used gene co-expression information to make trans regulation more tractable. INGENE used genetically predicted expression in a gene’s co-expression partners to model a target gene. MODULE instead focused on variants associated with the collective behaviour of a gene network.

The researchers trained and tested these approaches with RNA sequencing and genetic datasets spanning six post-mortem human brain regions. They then integrated the resulting expression predictions with schizophrenia genetic data from 62 cohorts in the Psychiatric Genomics Consortium’s third wave of analysis.

This is computational biology rather than an AI system diagnosing patients. The “AI” label is best understood as shorthand for data-driven modelling capable of finding patterns across large genetic and transcriptomic datasets. The models do not infer a person’s future diagnosis from a DNA sample, and the study does not propose a clinical tool for use today.

Why gene networks matter

The central scientific argument is that schizophrenia risk may be distributed across coordinated biological programmes rather than isolated genes. A regulatory variant can have a small effect at one point in a network, while the network’s combined activity may be more relevant to disease risk than any one component.

The team reported that incorporating trans-informed predictions improved coverage: it expanded the number of genes for which expression could be genetically predicted compared with cis-only approaches. In external testing using data from the Genotype-Tissue Expression project, the trans-aware models predicted more genes across the brain regions examined. The study also applied replication filters across datasets in an effort to reduce unstable trans-regulatory findings, which are typically harder to detect than local effects.

That broader coverage helped produce the list of 766 schizophrenia-associated genes. Importantly, “associated” has a specific meaning. It indicates that genetically predicted expression of a gene showed a statistical relationship with schizophrenia in the study design. It does not prove that altering that gene causes schizophrenia, nor that it would be safe or effective to target it with a medicine.

The distinction matters because psychiatric genetics has repeatedly shown that large numbers of small effects can be statistically robust while remaining difficult to translate into a discrete biological intervention. Associations can also reflect correlated genetic signals, shared regulation, or limitations in the available reference datasets.

From discovery to useful biology

The study’s near-term contribution is likely to be prioritisation. Researchers can use the expanded list to investigate whether particular genes converge on cell types, developmental periods, brain circuits or molecular pathways. Genes that repeatedly emerge across independent analyses may be candidates for laboratory experiments, including work in cellular models and more detailed studies of brain tissue.

The methods could also be useful beyond schizophrenia. Many common disorders involve non-coding genetic variation and complex gene regulation, including other psychiatric, neurological and metabolic conditions. A framework that adds distant regulatory effects may reveal signals missed by local models, provided it can be replicated across datasets and populations.

Yet scale is not the same as completeness. The analysis used brain tissue from deceased donors, which is indispensable for studying human brain expression but cannot capture all relevant life stages, environmental exposures, medication histories or disease states. Gene expression also differs by cell type, and bulk tissue measurements can obscure changes occurring in relatively rare cells.

Another limitation is that genetic prediction models depend on the datasets used to build them. Their performance and applicability can vary by brain region, ancestry and sample composition. Independent replication, functional experiments and more diverse reference panels will therefore be necessary before the new associations can be treated as settled biological targets.

A step toward precision, not a clinical breakthrough

Schizophrenia affects about 23 million people worldwide, according to the World Health Organization, and can involve hallucinations, delusions, disorganised thinking, cognitive difficulties and substantial social or occupational impairment. Effective care already includes antipsychotic medicines alongside psychological, family, social and rehabilitation support tailored to the person.

The new modelling does not change that standard of care. Its importance is upstream: it offers a more detailed hypothesis map for understanding how inherited risk may influence brain biology. If future experiments can distinguish causal mechanisms from statistical correlations, the work could eventually help identify more precise therapeutic pathways.

For now, the appropriate reading is measured. Advanced computational models have made the genetic landscape of schizophrenia more visible, especially the role of distant regulation and gene networks. They have not solved schizophrenia’s causes, produced a genetic diagnosis, or established a new treatment. But they may make the next generation of biological research more focused and testable.

Sources