A couple preparing for a family can accumulate a remarkable amount of information: genetic reports, laboratory results, reproductive history, previous treatment records, and a growing list of questions. Each record describes part of the picture. The challenge is understanding which findings matter individually, which matter together, and which questions the available information cannot yet answer.
The Couples Simulation Engine, or CSE, explores that shared biological context. Developed within the Nexus Health BioMatrix framework, it is designed to organize two contributing biological profiles, examine relationships between them, and produce research findings and proposed next steps for clinician review.
CSE is currently in beta. Its v3.0 clinical white paper describes a modular architecture for paired analysis, inheritance scenarios, reproductive context, quality checks, and reporting. This article follows that architecture from the first uploaded record to the final explanation, while distinguishing the questions a simulation can explore from the outcomes that require clinical evidence. [1]
1. Why two profiles change the question
Some health questions can be investigated primarily through an individual's records. Reproductive genetics introduces relationships that only become visible when two contributors are considered together.
For example, an individual may carry a disease-associated variant without having the corresponding recessive condition. The reproductive implications depend in part on whether the other contributor carries a relevant variant for the same condition. Reviewing two reports together can therefore raise a different question from reviewing either report alone.
Other information remains individual even within a paired analysis. A laboratory measurement belongs to one person, at a particular time, under particular circumstances. Combining two profiles should preserve those distinctions. A partner's reassuring result cannot cancel an abnormal finding in the other person's record.
CSE is built around this balance: preserve the individual evidence, then examine the relationships. Its central object, the CoupleDataModel, gives the workflow a shared structure without erasing who contributed each finding. The intended benefit is a clearer account of what is known about each person and what the combination may mean for a specific research question.
2. The model begins with records, roles, and context
The v3.0 design brings together three broad information layers: genetic data, laboratory findings, and intake history. Each layer contributes something different.
Genetic data supplies the variants actually represented in the submitted file. The white paper describes support for VCF, BCF, and an internal GSD format. A compatible file format is only the beginning: coverage, call quality, variant interpretation, and the purpose of the original test still matter. A consumer genotype file should not be treated as a comprehensive clinical carrier screen simply because it can be parsed.
Laboratory findings add dated observations about current biology. The proposed input categories include blood counts, selected hormone measurements, inflammatory markers, and nutrient-related findings. Which tests are appropriate is a clinical decision. The architecture's list of possible inputs is not a recommendation that everyone obtain every listed test.
Intake and history provide the context needed to interpret those records: medications, relevant diagnoses, reproductive history, symptoms, timing, and previous investigations. Two numerically identical hormone results may raise different questions when collected at different points in a cycle or under different treatment circumstances.
The v3.0 inheritance design primarily assumes an egg contributor and a sperm contributor, with additional reproductive context. Those are modeling roles, not a definition of who constitutes a family. Donor arrangements, gestational roles, and other family configurations need explicit representation rather than being forced into assumptions the current model does not support. [1, pp. 4–8, 12–17]
3. CoreDataBridge: building a reliable shared record
The first named module is CoreDataBridge. Its job is to establish the foundation on which later interpretations depend: connect the correct records, retain their provenance, and assemble the paired data model.
That means asking practical questions before asking biological ones. Are the files readable? Do they belong to the intended people? Are important fields missing? Does a laboratory value have usable units and a collection date? Does the genetic file contain sufficient information for the specific analysis being requested?
The white paper describes quality gates that can pause a workflow when inputs are incomplete or unsuitable. This is an important engineering property. If a later module depends on a finding that was never measured, producing a fluent answer would conceal the gap. An explicit pause makes the limitation visible and gives the reviewer a chance to resolve it.
CoreDataBridge also gives subsequent modules a consistent reference point. A finding should retain the same identity and meaning as it moves from the underlying record to an interpretation and then to a readable summary. That continuity is what makes a complex analysis inspectable. [1, pp. 14–18]
4. CarrierMapper: where paired genetics becomes concrete
CarrierMapper addresses one of the clearest reasons to examine genetic information together: inheritance. The design considers autosomal recessive, X-linked, and mitochondrial patterns, whose implications differ and require appropriate interpretation.
A simple autosomal recessive example helps explain the idea. If both genetic contributors are confirmed carriers for the same recessive condition, the standard inheritance model gives each pregnancy a 25 percent chance of an affected child, a 50 percent chance of an unaffected carrier, and a 25 percent chance of an unaffected child who is not a carrier. These probabilities apply independently to each pregnancy; they do not predict a fixed distribution among four children. MedlinePlus explains this inheritance pattern.
The word confirmed matters. A common variant associated with a small statistical effect is different from a pathogenic variant implicated in a Mendelian condition. The model should preserve the variant classification, evidence, and limitations of the original testing. Potentially significant findings may require clinical confirmation and genetic counseling before reproductive decisions are considered.
CarrierMapper's useful role is to make a relevant relationship visible and explain why it deserves review. A carrier finding does not make a person unhealthy, and a shared finding is not a judgment about a couple's suitability to have a family. The clinical conversation concerns evidence, choices, uncertainty, and the preferences of the people involved. [1, pp. 8, 14–18]
5. ReproPredictor: exploring possible combinations
ReproPredictor is the white paper's scenario-generation component. It describes constructing one to four hypothetical offspring profiles from the available parental information, then examining selected inherited combinations and their proposed implications.
A scenario can be a useful explanatory tool. It can show how a variant might be inherited, why a paired finding raises a question, or which uncertainty changes when a particular combination is considered. It can also help a reviewer compare assumptions that are otherwise difficult to hold in mind.
The number of scenarios must be interpreted correctly. Four simulated profiles do not exhaust the possible genetic combinations, predict four actual children, or describe birth order. A small set of examples is a way to explore a model. It is not a complete probability distribution over future offspring.
The distinction becomes especially important for complex traits and diseases. These can involve many genetic variants alongside environmental and other influences. A detailed-looking profile is still a model-based inference, with uncertainty that depends on the underlying evidence. The National Human Genome Research Institute describes polygenic risk scores as estimates derived from many variants, rather than certain predictions of a person's future health. Read NHGRI's overview of polygenic risk scores.
For CSE, the strongest use of a scenario is to make a question clearer: which inheritance pattern is being illustrated, what information supports it, and what would be needed to move from an illustration to a calibrated prediction? [1, pp. 8–11, 15–18]
6. FertilityAligner: adding the dimension of time
Genetic information is relatively stable, while reproductive measurements can depend strongly on timing and context. FertilityAligner is designed to bring that changing context into the paired analysis.
The white paper describes comparing available cycle information, hormone-related findings, and other reproductive observations. Its proposed use cases include organizing questions for preconception review and assisted-reproduction teams.
One practical contribution is temporal clarity. A report should distinguish a baseline measurement from a measurement collected during treatment, and a recent finding from an older one. If records describe different phases or circumstances, the model should retain that difference instead of treating the measurements as directly interchangeable.
The resulting analysis can help frame questions for a clinician: which results are comparable, what context is missing, and what needs review before any next step? The v3.0 architecture does not, by itself, establish a validated probability of conception, implantation, or live birth. Decisions about testing, medication, treatment timing, and embryo transfer remain with the reproductive care team. [1, pp. 9–11, 15–18, 23–24]
7. EpigeneticForecaster: naming the inference accurately
EpigeneticForecaster explores proposed relationships among genetic findings, methylation-related pathways, oxidative stress, and other biological context. It is one of the more ambitious parts of the architecture, and its outputs need particularly clear definitions.
Epigenetics concerns chemical modifications associated with DNA and its packaging that influence gene activity without changing the underlying sequence. Measuring genetic variants is therefore different from measuring an epigenetic state. A pathway-related laboratory marker also does not directly reveal a future child's epigenome. NHGRI provides a concise definition of epigenetics.
Within CSE, an inferred pathway-burden index should be presented as an inference produced by the model, with its inputs and assumptions visible. It should not be described as a direct measurement of an embryo or a diagnosis of a future child.
This creates a productive research question: can particular combinations of observed data support useful, reproducible hypotheses about biological processes? Answering it requires carefully defined outcomes and appropriate validation. The architecture provides a place to investigate those relationships; it does not settle them merely by assigning a score. [1, pp. 15–18]
8. NutrientHarmonizer: coordinated review, individual needs
NutrientHarmonizer is the proposed planning component for comparing nutrient-related findings and organizing individualized suggestions across both profiles. The useful concept is coordination: a paired report can explain where the questions overlap and where each person's situation differs.
A shared plan does not mean identical supplements or doses. Medical history, medications, laboratory findings, pregnancy considerations, and existing treatment plans can all change what a clinician considers appropriate. The report should expose those differences and keep proposed actions linked to the evidence for the individual concerned.
Common genetic findings also need evidence-based interpretation. For example, CDC guidance states that common MTHFR variants do not prevent folic acid processing and are not, by themselves, a reason to avoid folic acid. A genetic label should not automatically override established guidance or a clinician's assessment. Read the CDC guidance on MTHFR and folic acid.
CSE's proposed action plans are intended for clinician review. Their value depends on whether they help the care team understand the findings and make appropriate decisions, with room to modify or decline any suggestion. [1, pp. 16–18, 23–24]
9. The orchestration layer: keeping the analysis coherent
The six main modules rely on shared components that coordinate execution and review. These provide the engineering structure around the biological questions.
TaskMaster sequences the work so later modules receive the information they require. The white paper describes an ordered process in which upstream checks can stop downstream reporting. That prevents a missing prerequisite from being silently transformed into an apparently complete analysis.
MismatchMiner+ examines discrepancies between genetic context and laboratory findings, within and across the two profiles. A discrepancy can reveal an assumption worth revisiting. It is not automatically evidence that one source is wrong.
ResponderClassifier organizes internal response-related categories. Such categories are model constructs whose definitions and supporting evidence need to remain visible; a category name is not a clinical diagnosis.
QAFlagAgent identifies conditions that require attention, clarification, or a pause. A quality flag is most useful when it explains the triggering evidence, the affected part of the analysis, and what review is needed.
NarrativeSynth turns structured findings into reports for different readers. Its responsibility is to preserve meaning as technical material becomes readable. A tentative observation should remain tentative in the family-facing explanation, even when the prose becomes simpler. [1, pp. 16–20, 25–28]
10. Following one question through the engine
This example illustrates the workflow. It is not a patient case or a treatment recommendation.
Imagine two people arrive with genetic reports from different providers, laboratory results collected at different times, and a question about whether their records reveal anything they should discuss before pregnancy.
CoreDataBridge first establishes which records belong to which contributor and what each test actually covered. One file may contain broad sequencing data while the other contains a smaller set of genotype calls. The analysis must retain that difference rather than calling an unreported variant absent.
CarrierMapper then examines any supported inheritance questions. If a potentially significant finding appears in one report but the corresponding region was not adequately assessed in the other, the useful conclusion may be that the paired question remains unresolved. A clinician or genetic counselor can determine whether confirmation or additional testing is appropriate.
Meanwhile, FertilityAligner preserves the dates and context of the laboratory records. NutrientHarmonizer can organize relevant findings for review without assuming that both people need the same intervention. If a scenario is generated, ReproPredictor identifies the inheritance assumptions behind it.
The final report can be valuable even when it does not produce a dramatic prediction. It may clarify what is known, identify a missing piece that changes interpretation, and give the care team a better sequence of questions. That is a concrete outcome of good information architecture.
11. A report for clinicians and the people behind the records
The CSE white paper describes a four-part reporting pattern that separates evidence from explanation:
- Clinical Results: the relevant findings and structured outputs.
- Interpretation: the meaning assigned to those findings, including the model's reasoning and limitations.
- Nexi AI Notes: additional analytical context and review points generated within the system.
- Client Summary Notes: an accessible explanation for the people whose information is being reviewed.
That separation allows different readers to enter the report at the level they need. A clinician can inspect the evidence behind an interpretation. A family can understand what was reviewed and what questions remain. Both should encounter the same underlying meaning.
The document also describes HTML, PDF, and structured output ambitions. Those are reporting design choices, not proof of an integration with every clinical system. The essential feature is continuity: source information, analytical output, and readable explanation should remain connected through the review process. [1, pp. 19–20, 27–28]
12. What the beta evidence tells us
The v3.0 white paper reports 68 internal quality-assurance test cases and more than 100 simulation runs using a mixture of real and synthetic inputs during January 2024 through May 2025. These are development results reported by Nexus, rather than an independently audited clinical outcome study. [1, pp. 31–32]
Internal testing can answer meaningful engineering questions: does an invalid input trigger the intended block, does a module receive the right records, and does the report preserve a finding consistently? Those checks help establish whether the software behaves as designed.
Clinical prediction requires another layer of evidence. Agreement with a repeat analysis of genetic inputs does not establish that a system can accurately forecast a child's future health. A claim about prediction needs a clearly defined outcome, an appropriate comparison, and evaluation in people whose outcomes were not used to develop the model.
Clinical utility asks a further question: does using the system improve decisions or meaningful outcomes? The evidence presented in the v3.0 white paper does not establish improvements in pregnancy rates, miscarriage rates, live birth, or a child's later health. Those remain questions for appropriately designed studies.
This distinction is especially relevant to polygenic embryo screening. In its 2026 committee opinion, the American Society for Reproductive Medicine describes preimplantation genetic testing for polygenic disorders as an unproven technology that is not recommended for clinical use. CSE's research scenarios should not be mistaken for a validated embryo-selection service. Read the ASRM committee opinion.
13. The roadmap: broader models, stronger evidence
The white paper's future-development section proposes more quantitative polygenic modeling, dynamic cycle forecasting, richer family representations, and expanded support for donor and gestational configurations. It also discusses longer-term family and multi-offspring views.
These are roadmap ideas. Their appearance in a 2025 document does not confirm that they have been released, and a richer simulation does not automatically produce a better clinical prediction. Each expansion changes the questions the model must answer about data quality, interpretation, and validation.
Broader family representations are particularly significant. The person providing an egg, the person providing sperm, the person carrying a pregnancy, and the people parenting a child may occupy different roles. A useful system needs to represent the biological question accurately while respecting the family as it exists.
Longitudinal work introduces another opportunity: revisiting earlier interpretations as information changes. A new laboratory result, a clinically confirmed variant, or a revised scientific classification can change the meaning of an earlier report. Versioned records and clear reasoning make that change understandable.
The research direction is therefore both biological and methodological: improve the models, define their limits, and gather evidence that shows where they help. [1, pp. 29–30]
14. Making a complex conversation easier to have
The most compelling idea behind CSE is the possibility of bringing scattered information into a coherent conversation. Two people can arrive with separate records and leave with a clearer account of which questions belong to each individual, which emerge from their combined information, and which remain open.
Achieving that requires more than computational detail. It requires traceable inputs, transparent assumptions, careful language, and clinicians who can evaluate the findings in context. It also requires respect for the uncertainty and personal significance of reproductive decisions.
As CSE develops through beta, that is the standard its architecture invites: a simulation should help people and their care teams understand the evidence and ask better questions. Research findings and proposed treatment or action plans are shared for clinician review, with clinicians making care decisions.
To explore the underlying individual-analysis framework, read BioMatrix: Engineering an Integrated View of Human Biology. Individuals and families interested in participating can begin research intake or visit the Nexus Research FAQ. Registration begins the intake process; it does not by itself provide an immediate CSE simulation.
Source notes
Article type: an explanatory research feature based on Nexus technical documentation, with selected external references for scientific context. It is not a report of a newly conducted clinical trial.
[1] Nexus Health AI LLC. Biomatrix CSE v3.0 Clinical White Paper, July 2025, 34 pages. Primary sections: framework and use cases, pp. 4–11; inputs and architecture, pp. 12–20; applications and reporting, pp. 23–28; roadmap, pp. 29–30; internal testing, pp. 31–32.
Related Nexus documentation: John Whitman, BioMatrix v3.0 whitepaper, Q2 2025; John Whitman, Nexus Health Clinical Systems Division, BioMatrix v3.0 analysis: Backend Workflow Architecture & Report Generation Sequence, May 2025. These describe the individual-analysis framework on which CSE builds.
Scientific context: MedlinePlus: inheritance and risk; NHGRI: polygenic risk scores; NHGRI: epigenetics; CDC: MTHFR and folic acid; ASRM: 2026 opinion on polygenic preimplantation testing. These references support specific scientific context and do not validate the CSE product. Current beta status was confirmed by Nexus Health.