Depression is often treated as a single disorder in genetic studies, but its diagnostic criteria encompass a broad collection of experiences: low mood, loss of interest, problems with concentration, appetite and weight changes, altered sleep, fatigue, agitation and slowed movement. These symptoms do not necessarily share identical biological origins. In particular, hypersomnia and weight gain may represent a different neurobiological profile from insomnia and weight loss, a pattern sometimes associated with immunometabolic or “reversed-neurovegetative” depression. The new work therefore examined symptoms individually rather than assuming that every feature reflects one common depression factor. Its central question was whether biological pathways previously nominated in a mouse experiment involving nicotinamide mononucleotide, or NMN, would show symptom-specific associations with genetically predicted gene expression in the human brain.

To investigate that question, Ngo Cheung reconstructed pathway-level statistics from publicly available Summary-level PrediXcan, or S-PrediXcan, results. S-PrediXcan is a transcriptome-wide association method that uses genetic variants to predict how strongly genes are likely to be expressed in a tissue, then tests whether those genetically predicted expression levels are associated with a trait. In this case, the available gene-level Z scores covered 12 depression symptoms across multiple brain tissues. A Z score expresses the strength and direction of an association in standardized units: positive values indicate that higher genetically predicted expression tends to track with the symptom, while negative values indicate the opposite pattern. The researcher combined evidence across tissues and aggregated genes into Kyoto Encyclopedia of Genes and Genomes, or KEGG, pathways.

The most robust signal involved endocytosis and hypersomnia. The pathway produced a Stouffer Z score of approximately −5.52, a statistic generated by combining standardized evidence from many genes. Its permutation probability was 0.0025, and 194 genes contributed to the result. The association was consistent across six brain tissues and ranked first among 240 tested pathway comparisons. A size-matched union-null analysis, designed to compare the pathway with randomly assembled gene sets of similar size, produced a probability of 0.005. These safeguards are important because large pathways can appear significant simply by containing many genes. The result also survived checks designed to determine whether the association was distributed across the pathway rather than being driven by a narrow selection of genes with unusually large effects. According to the study, the signal aligned with the core direction of hypersomnia rather than with a small, contradictory subset of genes.

Endocytosis is a fundamental cellular transport process. It begins when a cell membrane folds inward and encloses molecules, receptors or membrane fragments in a small vesicle. The vesicle can then merge with compartments such as endosomes, where its contents are sorted, recycled or sent for degradation. In neurons, endocytosis is essential for recycling synaptic vesicles after neurotransmitter release, maintaining the balance of receptors at the cell surface and regulating communication between nerve cells. Disturbances in this machinery could, in principle, affect neuronal signaling, energy use or responses to external signals. But the study did not measure endocytosis directly in people with depression, and it did not establish whether the pathway is overactive or underactive in patients. The negative statistical association with hypersomnia means only that the direction of genetically predicted expression across the implicated genes was inversely related to the symptom in the analyzed data.

Other findings were more limited but added to the study’s picture of symptom diversity. Antigen processing and presentation, a biological process involved in displaying protein fragments to immune cells, showed a positive association with hypersomnia. However, the effect was small and heavily weighted toward TAPBP, a gene involved in loading peptides onto major histocompatibility complex class I molecules. This concentration means the result should not be interpreted as evidence that the entire immune pathway has a uniform relationship with excessive sleep. Rather, it identifies a specific gene and process for further investigation. The study’s authors treated the immune result as less definitive than the endocytosis signal, emphasizing the importance of distinguishing a pathway that is broadly supported from one that appears significant because of one influential component.

The analysis also found positive associations between psychomotor agitation and two gene sets: peroxisome and a vasopressin-labelled set. Peroxisomes are intracellular organelles that help break down fatty acids, manage reactive oxygen species and carry out other aspects of lipid and energy metabolism. In the agitation analysis, the signal was supported mainly by IDH2 and HSD17B4 rather than by HMGCL. IDH2 participates in mitochondrial metabolism and redox balance, while HSD17B4 has roles in peroxisomal fatty-acid processing. The vasopressin-labelled set, meanwhile, was driven more strongly by dynein and genes in the CREB3 family than by AVP itself. Dynein is a motor protein complex that transports cargo along microtubules, and CREB3-family proteins regulate gene expression in response to cellular stress and secretory demands. These results illustrate why pathway labels alone can be deceptive: a set named after a hormone or organelle may owe its statistical association to genes with quite different cellular functions.

One especially dramatic gene-level contrast involved HMGCL, which showed an extreme difference between weight gain and weight loss. HMGCL encodes an enzyme involved in ketone-body production, linking it to energy metabolism during periods when carbohydrate availability is low. Such a sharp contrast might appear to connect HMGCL directly to the peroxisome–agitation association, but the study found that it did not. This distinction is central to interpreting modern genetic pathway analyses. A gene can show a compelling association with one symptom while contributing little or nothing to a separate pathway signal involving another symptom. Conversely, a pathway can appear important because several modestly associated genes point in the same direction, even when no single gene dominates. The researcher therefore used leave-one-gene-out tests, bootstrap confidence intervals and gene-class decomposition to examine how stable each result remained when individual contributors were removed.

The work is best viewed as a prioritization study rather than a discovery of a depression mechanism or a treatment target. Its data were derived from summary statistics and genetically predicted expression, not from direct measurements of RNA or endocytosis in the brains of people experiencing depression. Genetic prediction also does not capture every factor that controls gene activity, including medication, stress, sleep history, metabolic state, cell type and environmental exposures. The analysis cannot show that NMN improves depression, that NMN changes human brain pathways, or that patients with a particular symptom profile would respond to NMN or any other intervention. Nor does it establish biological subtypes of depression. The NMN connection comes from the origin of the nominated KEGG gene sets in a mouse aging experiment; it is not evidence of NMN responsiveness in humans.

Even with those limitations, the findings could influence how future depression research is designed. Instead of asking whether a gene is associated with “depression” in the broadest possible sense, investigators may test whether it is linked specifically to hypersomnia, insomnia, weight gain, weight loss, agitation or psychomotor slowing. The endocytosis–hypersomnia association could be examined in larger cohorts, directly measured in relevant neuronal and glial cell types, and tested using colocalization analyses to determine whether the same genetic variants influence both gene expression and the symptom. Researchers could also investigate whether the signal varies by brain region, ancestry, age, sex or metabolic status. For now, the study’s most consequential message is methodological as much as biological: depression’s symptoms may carry distinct genetic signatures, and a pathway label is only the beginning of the explanation—not the explanation itself.

Single-cell RNA sequencing (scRNA-seq) has rapidly evolved into an indispensable tool for dissecting the molecular signatures of individual cell populations within complex brain structures. This technology facilitates the identification of cell type-specific transcriptional profiles, offering granular insight into how gene expression varies not only across different cells but also between pathological and healthy states. Its application in neuropsychiatric disorders is particularly poignant, given the heterogeneity of neuronal and glial subtypes implicated in these illnesses. Importantly, scRNA-seq also enables the reconstruction of pseudo-time trajectories, allowing scientists to track the dynamic progression of transcriptional changes, potentially illuminating the temporal sequence of cellular dysfunctions leading to disease phenotypes.

In tandem with transcriptomics, chromatin accessibility assays such as ATAC-seq (Assay for Transposase Accessible Chromatin using sequencing) have emerged as vital complements that contextualize gene expression within the framework of chromatin remodeling. This technique interrogates higher-order chromatin structures and identifies regulatory elements such as enhancers and promoters that govern gene activity. When integrated with single-cell ATAC-seq (scATAC-seq) and scRNA-seq, researchers can link epigenomic regulation directly to transcriptomic outputs at the single-cell level. This multimodal approach is revolutionizing our understanding of the epigenetic mechanisms driving cell type-specific transcriptional variations in brain tissues obtained postmortem, providing a more comprehensive depiction of gene regulation in MDD and suicide.

Beyond sequencing-based approaches, spatial transcriptomics has pioneered a spatially resolved dimension to molecular brain research. By placing thin tissue cryosections on barcoded slides, spatial transcriptomics captures the locality of mRNA molecules within their native tissue architecture. The spatial barcodes incorporated during cDNA synthesis allow each sequenced transcript to be mapped back to its precise anatomical coordinates. This breakthrough technology circumvents the limitations of dissociated single-cell methods by preserving the spatial context necessary for understanding intercellular interactions and microenvironment influences, which are crucial in brain circuits underlying mood regulation and suicidal ideation. Although nascent in human brain studies, preliminary investigations have mapped the cellular diversity in critical areas such as the hippocampus and prefrontal cortex, shedding light on their layered cytoarchitecture and potential disruption in mental illness.

Complementing transcriptomics, proteomics interrogates the functional end products of gene expression—the proteome. Protein expression and post-translational modifications fluctuate dynamically in response to cellular stresses and environmental cues, making proteomics indispensable for understanding pathogenesis at the molecular and systems levels. Traditional bulk proteomic analyses have benefitted from advances in mass spectrometry and microarrays, capturing protein abundance across tissues. More recently, single-cell mass spectrometry (scMS) has surged as an innovative technology capable of analyzing proteins and their modifications in individual cells without the constraints of affinity reagents, thus unlocking complex multimodal data layers that bridge gene expression and cellular phenotype. Techniques such as DBiT-seq (Deterministic Barcoding in Tissue for spatial omics sequencing) further augment proteomic spatial resolution, enabling focused analyses of brain regions implicated in MDD and suicidal behaviors.

These novel omics modalities generate prodigious amounts of complex, high-dimensional data demanding sophisticated computational tools for interpretation. Artificial intelligence has stepped into this arena as a formidable ally to biomedical researchers. Machine learning, a branch of AI grounded in pattern recognition, excels in feature extraction, model construction, and validation. Algorithms such as random forest and support vector machines (SVM) have become mainstays in neuropsychiatric biomarker discovery and classification, successfully distinguishing disease states with high accuracy. For instance, applying a random deep forest combined with leave-one-out cross-validation to blood samples harnessing both differentially expressed genes and methylated CpG sites remarkably achieved over 90% accuracy in differentiating suicidal from non-suicidal individuals with depression, underscoring AI’s potential for precision psychiatry.

Deep learning, a more complex AI subset inspired by neural networks, offers automated, multi-layered feature learning capable of navigating the vast complexity and heterogeneity typical of mental health datasets. Utilizing transcriptome, genomic variants (SNPs), and three-dimensional chromatin conformation data (Hi-C), deep learning integrated with weighted gene co-expression network analysis (WGCNA) has elucidated shared immune and synaptic gene networks across bipolar disorder and schizophrenia, unraveling common biological threads underpinning distinct psychiatric conditions. The power of deep learning is monumental in integrating multi-omics with neuroimaging and clinical records, amplifying the prospects for early diagnosis, risk stratification, and personalized therapeutic interventions in psychiatry.

The burgeoning availability of diverse data modalities—from clinical narratives and electronic health records (EHR) to neuroimaging and molecular profiles—places deep learning at the forefront of mental health innovation. AI models trained on these heterogeneous datasets are uncovering subtle language markers and behavioral signals predictive of depressive and suicidal tendencies. Social media textual analysis using deep neural networks has penetrated the barrier between lay communication and clinical symptoms, detecting linguistic patterns indicative of depression with considerable accuracy. Likewise, mining EHR data through these algorithms enhances suicide risk prediction by detecting complex patterns hidden within clinical notes, signifying a paradigm shift in mental health screening and intervention.

Natural language processing (NLP), an AI technique specializing in transforming unstructured text into analyzable, structured data, supports clinicians in deciphering patient speech and behavioral patterns during psychiatric evaluations. This bridges human linguistic nuances and computational analytics, crafting datasets that underpin algorithmic diagnoses and monitoring. Voice analysis, increasingly coupled with NLP, can detect subtleties in tone, pitch, and speech patterns reflective of mental states, augmenting objective assessments in psychiatric practice and offering new avenues to track disease progression or therapeutic response.

Together, these advances carve a path toward a more mechanistic and individualized understanding of depression and suicide. The combined use of advanced omics technologies with AI-driven analyses heralds a new era where the multi-layered complexity of brain function and dysfunction can be systematically deconvolved. This integrated approach promises to transform psychiatric diagnostics from symptom-based assessments to biologically grounded precision medicine, potentially reducing the stigma and enhancing treatment efficacy.

While challenges remain—such as the need for larger, well-characterized cohorts, improved data harmonization, and ethical considerations around patient data privacy—the momentum in omics and AI research underscores an optimistic outlook. The integration of spatially resolved transcriptomics, epigenomic profiling, and proteomics with machine learning and deep learning algorithms is refining the neurobiological models of MDD and suicidal behavior, enabling the uncovering of novel biomarkers and therapeutic targets.

As these innovative approaches continue to mature, their ability to map the cellular and molecular ecosystems of the brain with spatial, temporal, and functional precision will deepen our knowledge. This may herald the development of diagnostic tools capable of predicting suicide risk with high fidelity or uncovering druggable molecular networks that could be modulated for effective intervention. Ultimately, the fusion of omics and AI embodies a new frontier in mental health research, turning complex biological data into actionable clinical insights.

The implications stretch beyond depression and suicide, offering a template for tackling other neuropsychiatric disorders where heterogeneity and complexity have impeded progress. By embracing these technologies, the scientific community moves closer to fulfilling the long-standing promise of personalized psychiatry—where treatments are tailored to molecular signatures, trajectories are predicted before clinical deterioration, and suicide becomes a preventable tragedy.

In summary, the synergy between omics technologies and artificial intelligence is reshaping the terrain of depression and suicide research. Innovations in single-cell sequencing and spatial transcriptomics illuminate cell-type-specific changes within key brain regions. Proteomics and scMS expand this understanding to protein landscapes. Meanwhile, AI-driven pattern recognition and deep learning models extract meaningful insights from voluminous datasets, driving forward early detection, mechanistic understanding, and precision intervention strategies. This confluence of cutting-edge biology and computational prowess marks one of the most exciting frontiers in neurology and psychiatry, destined to transform the clinical management of one of humanity’s most pressing mental health challenges.

Subject of Research:
Recent developments in omics technologies and artificial intelligence applied to understanding depression and suicidal behavior.

Article Title:
Recent developments in omics studies and artificial intelligence in depression and suicide.

Article References:
Wang, Q., Dwivedi, Y. Recent developments in omics studies and artificial intelligence in depression and suicide.
Transl Psychiatry 15, 275 (2025). https://doi.org/10.1038/s41398-025-03497-y

Image Credits:
AI Generated

DOI:
https://doi.org/10.1038/s41398-025-03497-y