In an ambitious stride toward unraveling the complexities of psychosis and its prodromal stages, researchers involved in The Accelerating Medicines Partnership® Schizophrenia Program (AMP®SCZ) are pioneering a multifaceted approach to identify predictive biomarkers. This initiative aims to revolutionize early detection by integrating cutting-edge genomic assays, advanced proteomic technologies, and precise hormonal measurements, all underpinned by robust computational models. The ultimate goal: to create a clinically actionable risk calculator that could transform psychiatric diagnostics and intervention strategies.
At the core of this innovative endeavor lies the exploitation of polygenic scores, which amalgamate the cumulative effect of numerous genetic variants associated not only with psychosis but a spectrum of psychiatric disorders including schizophrenia, bipolar disorder, depression, attention deficit hyperactivity disorder (ADHD), and autism. By leveraging polygenic risk across these overlapping disorders, researchers hope to enhance predictive accuracy beyond traditional clinical assessments. Central to this genomic profiling is the adoption of the Blended Genome Exome (BGE) assay, a cost-effective sequencing technique designed to capture a broad swath of genetic variation.
The BGE assay distinguishes itself by balancing depth and breadth: it sequences the exome—the protein-coding portion of the genome—with high coverage around 30X, ensuring sensitive detection of rare, potentially pathogenic variants. Simultaneously, it surveys the remaining 98% of the genome at a low coverage between 1X and 3X, a calibrated depth optimized to detect common variants across diverse ancestries. This dual-faceted approach facilitates not only the detection of single nucleotide variants but also important structural alterations such as copy number variants, which have been implicated in psychiatric conditions.
Ensuring data integrity and reliability in such extensive sequencing endeavors is paramount. The AMP®SCZ team implements rigorous quality control (QC) measures encompassing multiple parameters: coverage metrics for both the exome and whole genome, per sample and variant call rates, contamination indices, and indicators of library preparation artifacts such as chimeric reads. Ancestry-specific filters based on median absolute deviations and genetic quality metrics like transition/transversion ratios and heterozygosity ensure outlier exclusion. Intriguingly, samples exhibiting discordance between reported biological sex and genetically inferred sex are systematically excluded to maintain dataset fidelity. Subsequent imputation of sequencing data, leveraging reference population genotypes, enables comprehensive polygenic score calculation rooted in large-scale genome-wide association studies (GWAS).
Beyond the genetic landscape, the study rigorously incorporates endocrine biomarkers, specifically salivary cortisol, owing to its well-documented involvement in stress-related psychosis risk. The collection protocol involves sampling saliva at three discrete time points over a two-hour window, with immediate freezing to preserve sample integrity. Cortisol quantification utilizes the Salimetrics enzyme-linked immunosorbent assay (ELISA) platform, performed consistently across two separate facilities using assays from the same manufacturer lot to preclude batch effects. Recognizing diurnal variations inherent to cortisol physiology, the measured values are adjusted accordingly along with other confounding variables. The averaged adjusted cortisol level is subsequently integrated into the risk prediction framework, enriching the biological dimensions of causality and prediction.
Proteomics, an indispensable pillar in the quest for biomarker discovery, is meticulously targeted to include proteins implicated in neuroinflammation, complement activation, coagulation pathways, and oxidative stress—processes intimately linked to psychotic pathophysiology. Furthermore, the analysis encompasses brain-derived blood proteins and molecules encoded by genes associated with schizophrenia susceptibility, alongside a comprehensive survey of blood-secreted proteins. To balance cost and analytical coverage, the project is evaluating two leading proteomic platforms: Olink and SomaScan, both commercial multiplex technologies with robust validation in clinical proteomics.
The Olink platform is predicated on a proximity extension assay that employs pairs of antibodies anchored to unique DNA oligonucleotides. Upon binding to the protein target, these oligonucleotides hybridize to form a DNA duplex, which is then amplified and quantified via sensitive real-time PCR methods. Contrastingly, SomaScan technology harnesses chemically modified, fluorescently labeled single-stranded DNA aptamers designed to bind target proteins with exceptional specificity and sensitivity. Both platforms can assay thousands of proteins spanning a dynamic concentration range congruent with plasma proteome complexity, and exhibit impressive reproducibility with minimal cross-reactivity—a critical factor for unbiased biomarker quantitation.
Data preprocessing takes a methodical path where raw proteomic measurements undergo manufacturer-recommended quality control filtering. The expectation is that most proteins demonstrate stable expression between baseline and two-month follow-up samples, enabling use of coefficient of variation distributions as a proxy for platform reproducibility. Analytical tools such as principal component analysis and Grubbs’s test help detect outliers, while assessments of hemolysis indicators, sample processing intervals, and physiological confounders like body mass index ensure biological validity of protein signals. Additional QC measures include assays targeting proteins sensitive to ex vivo blood cell or platelet activation—events known to artifactually inflate certain biomarker levels—thus safeguarding data authenticity.
The integration of these diverse biomarker modalities into clinically predictive models demands advanced computational strategies. Machine learning (ML) approaches, recognized for their potent pattern recognition and variable combination capabilities surpassing univariate analyses, are central to model development. However, the research team is acutely aware of the pitfalls posed by overfitting, where an algorithm may perform exquisitely on training datasets yet falter upon independent validation. To mitigate this, the methodological framework incorporates algorithmic safeguards explicitly designed to restrain overfitting tendencies.
Internal validation techniques such as cross-validation and bootstrap resampling provide repeated estimates of model generalizability by partitioning and re-sampling the dataset in multiple iterations. Such resampling mimics external population sampling variability, allowing robust performance metrics to be gleaned prior to prospective testing. Moreover, permutation testing serves as a critical statistical check by randomizing outcome labels and recalculating model accuracy thousands of times; if models outperform these null permutations, it strongly suggests true predictive signal rather than artifacts or chance correlations.
Through these multilayered strategies, the AMP®SCZ program anticipates constructing multivariable classifiers that forecast psychosis onset and related outcomes with unprecedented precision. The combinatory power of genetic, proteomic, and hormonal biomarkers, interpreted through sophisticated machine learning, promises to surmount current diagnostic limitations and elucidate the biological underpinnings of psychotic disorders.
The potential clinical impact of such predictive tools is transformative. Early identification of individuals at highest risk could enable targeted preventive interventions, optimized treatment selection, and personalized monitoring, fundamentally reshaping clinical psychiatry. Furthermore, the large-scale genetic and proteomic datasets generated will contribute to broader scientific understanding, facilitating discovery of novel therapeutic targets and pathways implicated in psychosis.
Future directions include validating these multivariate risk classifiers across diverse populations to ensure generalizability, refining biomarker panels for maximal cost-effectiveness and clinical utility, and integrating environmental and lifestyle data for comprehensive risk modeling. The AMP®SCZ’s commitment to open science and collaborative research accelerates this translational trajectory, setting a new benchmark for psychiatric biomarker development.
In sum, the Accelerating Medicines Partnership® Schizophrenia Program’s multifaceted biomarker exploration epitomizes a paradigm shift toward precision psychiatry. By harnessing the power of genomic sequencing, proteomic profiling, cortisol dynamics, and cutting-edge machine learning, this initiative seeks not only to predict psychosis risk with high fidelity but also to uncover the mechanistic biology that drives this enigmatic illness. The convergence of these technologies signals a hopeful horizon where mental illnesses are detected earlier, understood more deeply, and treated more effectively.
—
Subject of Research:
Biomarker development and risk prediction in psychosis through integration of genomic, proteomic, and endocrine data.
Article Title:
Body fluid biomarkers and psychosis risk in The Accelerating Medicines Partnership® Schizophrenia Program: design considerations.
Article References:
Perkins, D.O., Jeffries, C.D., Clark, S.R. et al. Body fluid biomarkers and psychosis risk in The Accelerating Medicines Partnership® Schizophrenia Program: design considerations. Schizophr 11, 78 (2025). https://doi.org/10.1038/s41537-025-00610-4
Image Credits: AI Generated
