SPECTRA ISO ARPA-H-SOL-26-163.pdf

Innovative Solutions Opening (ISO)

Systems for Phenotypic Evaluation, Clinical Trajectories, Response, and Agency (SPECTRA)

Proactive Health Office (PHO)

Advanced Research Projects Agency for Health

ARPA-H-SOL-26-163

September 17, 2026

Contents

PART I: OVERVIEW AND PROGRAM DESCRIPTION

1.1 Program Summary

1.2 Problem Statement

1.3 Program Vision and Impact

PART II: TECHNICAL AREAS AND REQUIREMENTS

2.1 Solicited Components: SPECTRA Technical Areas

2.1.1 Technical Area 1 (TA1): Multi-Factor Analysis and Trajectory Identification

2.1.2 Technical Area 2 (TA2): Population-Scale Modeling and Data Integration

2.1.3 Technical Area 3 (TA3): Precision Intervention and Trajectory Testing

2.1.4 Technical Area 4 (TA4): Technologies for Communication, Quality of Life, and Embedded Assessment

2.2 Shared Non-Solicited Program Resources

2.2.1 Program-wide Data Coordination

2.2.2 Community Engagement and Co-Design

2.2.3 Responsible-Use Controls

2.2.4 Independent Evaluation (IV&V)

2.2.5 Transition Support

PART III: PROGRAM STRUCTURE AND SCHEDULE

3.1 Program Metrics

PART IV: PROPOSAL REQUIREMENTS

4.1 Eligibility

4.2 Award Instruments

4.3 Solution Summaries (Pre-Proposal Submissions)

4.4 Proposal Requirements

4.5 Full Proposal Submission Requirements

PART V: EVALUATION CRITERIA AND SELECTION

5.1 Solution Summary Evaluation

5.2 Proposal Evaluation

5.3 Use of Large Language Models (LLMs)

5.4 Selection

PART VI: GENERAL INFORMATION

6.1 Research Security

6.2 Intellectual Property

6.3 Publication and Dissemination

6.4 Disclaimers

6.5 Questions and Answers

APPENDIX

INNOVATIVE SOLUTIONS OPENING (ISO)

Federal Agency: Advanced Research Projects Agency for Health (ARPA-H) Proactive Health Office

(PHO)

Program Title: Systems for Phenotypic Evaluation, Clinical Trajectories, Response, and Agency

(SPECTRA)

Announcement Type: Solicitation

ISO Solicitation Number: ARPA-H-SOL-26-163

Issue Date: September 17, 2026

Solution Summary Submission Deadline:

TA 4: October 18, 2026 at 5pm ET (submission portal opens Oct 4, 2026)

TA 1 and TA 2: December 2, 2026 at 5pm ET (submission portal opens Nov 18, 2026)

• Anticipated ISO Amendment late Oct

TA 3: January 18, 2027 at 5pm ET (submission portal opens Jan 4, 2027)

• Anticipated ISO Amendment early Dec

Proposal deadline information will be forthcoming.

Where to Submit:

Solution Summaries: https://solutions.arpa-h.gov

Questions: https://solutions.arpa-h.gov/Ask-A-Question/

Registration is required to submit documents via the ARPA-H Solutions site. Late submissions may not be considered.

Program Manager: Russell Shilling, PhD

Point of Contact: SPECTRA@arpa-h.gov

Award Instrument: Other Transaction (OT)

Resource Sharing: Resource sharing is encouraged but not required.

https://solutions.arpa-h.gov/ https://solutions.arpa-h.gov/Ask-A-Question/

PART I: OVERVIEW AND PROGRAM DESCRIPTION

1.1 Program Summary

Autistic people are valued members of their families and communities across the United States, contributing strengths, skills, and perspectives to family life, work, science, the arts, and civic life.

Yet, there is much we do not understand about autism. One in 31 American children has been diagnosed with autism spectrum disorder (ASD), and autism affects individuals, families, and caregivers across the full lifespan. Co-occurring health conditions are common: for example, roughly 30 percent of autistic people have co-occurring intellectual disability, sleep problems affect approximately 60 percent of children and adolescents, gastrointestinal symptoms affect at least one third, epilepsy affects approximately 1 in 10, and anxiety disorders affect approximately 1 in 5, with higher estimates reported in children and adolescents. Converging evidence indicates that autism arises from the interplay of many factors, including genetic, biological, behavioral, environmental, and iatrogenic (medical) influences, rather than from any single cause. How these factors combine to produce autism and its co-occurring conditions, and why they converge most profoundly in some individuals, is not yet understood. Existing research has produced compelling findings, but it has largely examined these factors in isolation and has not addressed the complexity of their nonlinear interactions or how those interactions unfold as trajectories across the lifespan. As a result, current care often leaves families without timely answers about cause, trajectory, communication, co-occurring conditions, and likely response to intervention.

The Advanced Research Projects Agency for Health (ARPA-H) seeks submissions for the Systems for Phenotypic Evaluation, Clinical Trajectories, Response, and Agency (SPECTRA) program.

SPECTRA will identify and act on autism trajectories, the evolving developmental paths individuals follow across the lifespan from the prenatal period through aging, spanning autism characteristics, co-occurring conditions, and functional outcomes together with the genetic, biological, behavioral, iatrogenic (medical influences), environmental factors and their nonlinear interactions. Rather than examining these factors in isolation, SPECTRA will examine how they combine and interact, including their nonlinear interactions, to produce autism and its co-occurring conditions over time.

Because different trajectories arise from different combinations of biomarkers and factors, SPECTRA encourages examining them in combination, whether assembled within one team or through partnership. By resolving these trajectories, SPECTRA aims to move autism care from one-size-fits-all management to biologically and behaviorally informed intervention: to explain not only what a person experiences but why, to enable earlier identification of autism and co-occurring conditions, and to guide personalized support and treatment that reduce the impact of co-occurring conditions and improve quality of life across every life stage. Understanding how these factors converge also opens opportunities to intervene before the most disabling features take hold. Where a modifiable combination of influences drives a trajectory toward regression, medical crisis, or profound impairment, resolving it early could reduce the severity of these outcomes and, in some cases, prevent them.

SPECTRA will also develop technologies that serve two purposes at once: extending reliable assessment beyond the clinic while improving everyday communication, function, and quality of life. The same tools people use to live more fully will also generate the data that drive measurement and personalized care.

SPECTRA supports and respects autistic people and their families. The program works to understand developmental trajectories and to improve health, communication, function, and quality of life across the full range of the spectrum. Consultation with the community and co-design where applicable will be key components of the program. SPECTRA gives particular attention to those with the highest support needs, including people with profound autism, who are most often excluded from research and least well served by current tools.

A note on scope. In this solicitation, medical exposures refers to therapeutic, prophylactic, and procedural exposures encountered in the course of clinical care across the lifespan, assessed with dose and timing resolution, both beneficial and iatrogenic. Immune refers to infection, fever, inflammatory and autoimmune processes, and immune activation from any source, in the parent or the individual. SPECTRA prejudges no factor and excludes none from examination.

A note on language. In alignment with major autism advocacy organizations, ARPA-H acknowledges that preferences regarding autism terminology vary within the autism community.

Some individuals prefer person-first language (for example, "person with autism"), while others prefer identity-first language (for example, "autistic person"). ARPA-H does not endorse one convention over the other and may use both forms interchangeably across its publications, while respecting individuals' stated language preferences. Where this document uses the term "profound autism," it does so descriptively, to refer to autistic people with high support needs, including those with co-occurring intellectual disability and minimal or no reliable speech. The term is not a formal diagnostic category, and ARPA-H uses it to make explicit that SPECTRA includes this population, not to define a subtype.

The Gap

Current research rarely examines the complex, nonlinear interactions among the genetic, biological, behavioral, environmental, and iatrogenic (medical) influences that shape autism, and it seldom connects them to the everyday outcomes families care about across the lifespan.

Assessment also happens largely in the clinic rather than in the real-world settings where people live, learn, and work, which limits what can be measured and when. ARPA-H believes this gap is now bridgeable. Converging advances in artificial intelligence and machine learning (AI/ML), genetic and biological markers, biosensing wearables, and voice and speech technology make it possible to resolve these interactions, match risk to phenotype, identify autism sooner, and deliver precision therapies beyond the clinic.

SPECTRA will pursue four Technical Areas (TAs) of research and development. TA1, Multi-Factor Analysis and Trajectory Identification, uses deep phenotyping to identify, measure, and model how the many contributing factors combine, and to define the trajectories and the harmful combinations that drive negative outcomes. TA2, Population-Scale Modeling and Data Integration, builds durable, secure national infrastructure that integrates and characterizes data across populations, so trajectories can be assessed for representativeness and people can be followed as they age. TA3, Precision Intervention and Trajectory Testing, tests in rigorous trials whether acting on these trajectories changes outcomes, including the co-occurring conditions that drive the most harm. TA4, Technologies for Communication, Quality of Life, and Embedded Assessment, develops tools such as wearables, communication and speech interfaces, eye tracking, low-channel EEG, and camera-based movement analysis that improve everyday communication and function while generating the real-world measurements that assessment now lacks.

TA1 and TA4 are designed to answer each other's central question. Today neither is sufficient on its own: the measurements needed to observe trajectories in daily life do not yet exist, and the models needed to interpret those measurements are nonlinear and incomplete. SPECTRA breaks this dependency through deliberate co-development. TA1 defines and models the trajectories worth measuring, which tells TA4 what to build; TA4 develops novel real-world measurements, which TA1 incorporates and tests against its models. In effect, each supplies the other's answer key. Each iteration sharpens both, until the program converges on a functional definition of the causal trajectories and outcomes that matter and the measurements needed to assess them. These same measures will be used in TA3 to assess effectiveness of therapeutic approaches.

A separate, non-solicited program element will support data coordination, independent verification and validation, community co-design, responsible-use controls, and transition support (see Section 2.2 for more on the non-solicited program elements).

1.2 Problem Statement

Current approaches to improving diagnostic capabilities and precision care for individuals affected by autism fall short in five significant ways:

First, diagnosis comes too late. Autism can often be reliably identified only well after the earliest signs emerge, and in many cases diagnosis is delayed by years, particularly for girls, those with subtler presentations, and those with limited access to specialty diagnostic care. Because the earliest years of life represent a period of exceptional neurodevelopmental plasticity, late identification narrows the window in which intervention can meaningfully improve developmental trajectories. Delay also allows secondary and comorbid conditions, such as anxiety, gastrointestinal disorders, epilepsy, and self-injurious behavior, to take hold and compound, worsening long-term outcomes that earlier action might have prevented or mitigated.

Second, environmental and developmental exposures are not integrated with host biology.

Autism does not follow a single cause or pathway; it likely reflects a dynamic interplay among an individual’s biology, genetics, the developing brain and body, and environmental and developmental exposures accumulated over time, combining differently from one person to the next. Yet current approaches largely evaluate these dimensions in isolation, if at all. Current methodologies rarely capture how exposures interact with an individual’s biology across development, and as a result cannot represent the full, longitudinal context that shapes each person’s phenotype. This leaves clinicians and researchers without a mechanistic picture of why a given trajectory unfolds as it does.

Third, the data that would reveal trajectories are siloed. The multi-modal data needed to see how autism unfolds are fragmented across health systems, research cohorts, and administrative records, held in incompatible formats and under separate governance. Because these sources cannot be analyzed together, no single view captures a representative population at sufficient scale, and patterns that would emerge only across combined data remain invisible. This leaves the field without the population-level foundation that trajectory identification and precision care require.

Fourth, data do not reliably translate into action. Even where rich developmental data exist, existing prediction models do not adequately capture the nonlinear, branching, and highly individual nature of developmental processes. They tend to describe risk or category membership rather than forecast an individual’s trajectory, and they offer little guidance on which intervention is likely to help which individual at which point in time. Consequently, data can characterize a person without informing what should actually be done to improve their outcomes.

Fifth, current assessments fail many of the people who most need them. Most instruments were designed around a narrow profile using a structured assessment in the clinic, and they capture only a thin slice of function and behavior, usually verbal and behavioral performance in a single setting, rather than the full range of function, self-regulation, sensory and behavioral needs, health and safety, learning, social participation, and communication. These assessments also conflate distinct domains: they conflate motor speech, cognition, and communicative competence, so that a difference in one domain is read as a deficit in another. This falls hardest on those with the highest support needs. An estimated 25 to 30 percent of autistic people are minimally verbal or nonspeaking, and standard tools were not designed for them; the same tools also underserve people with intellectual disability, co-occurring conditions, and atypical presentations. This mismatch limits both the ecological validity of assessment and the durability and relevance of intervention.

Together, these limitations leave autism care without a robust path from early, ecologically valid, multidimensional measurement to individualized prediction and precision intervention delivered when and where they can do the most good. SPECTRA is motivated by the hypothesis that autism can be understood as a dynamic developmental trajectory, one whose phenotypic features, biological and environmental drivers, and responses to intervention can be measured early and continuously in real-world settings, integrated across host biology and exposure, and used to forecast and adaptively shape outcomes toward greater health and agency.

1.3 Program Vision and Impact

If successful, SPECTRA will deliver personalized, evidence-based care and support for people with autism across the lifespan. It will improve understanding of the causal and modifiable influences that shape developmental, health, and functional trajectories, and translate that understanding into biologically and behaviorally grounded tools for earlier identification and precision intervention. The program moves autism care beyond a single diagnostic label toward trajectory-guided treatment that improves communication, health, independence, autonomy, and quality of life while preventing avoidable morbidity.

SPECTRA will deliver the following outcomes:

• Evidence on the factors and phenotypes that shape individual trajectories and which of them are potentially modifiable. This will enable care to target what drives outcomes rather than the diagnostic label alone (TA1).

• Trajectory-guided precision intervention that tests what works for whom, including when to safely start, adjust, or stop treatment. This will replace the trial-and-error approach that is the current standard of care (TA3).

• Assessment and diagnostics that work in everyday settings, including home, school, clinic, and community, measuring communication, function, and treatment response during daily life.

This will allow the ability to intervene with issues such as distress, pain, sensory overload, sleep disruption, and seizures before they become crises (TA4).

• Communication and functional supports that preserve the person's intent, autonomy, and agency across the full range of support needs (TA4).

• Durable, privacy-preserving learning systems that improve tools and care models as longitudinal data accumulate, without forcing personal data into central custody (TA2).

• Family and caregiver quality of life treated as an explicit outcome, not an assumed consequence of care (cross-cutting).

Together, these outcomes will identify treatable conditions earlier, including but not limited to sleep disorders, seizures, gastrointestinal symptoms, anxiety, and pain, and support earlier action to prevent avoidable morbidity, regression, crisis, and loss of function across the lifespan.

Evidence Tiers:

All performers are bound to common, clear, agreed-upon language denoting the strength of evidence for causality for all publications, materials, or public communication. All work supported by the program will adhere to a three-tier evidence standard, which the program will publish. For clarity, language about cause is reserved for the tier three evidence this program will generate.

Nothing at tier one or two may be described as driving, causing, or leading to an outcome.

Tier Definition Permitted language

I: Associated Observed in data, not replicated, confounding not excluded associated with, observed in

II: Replicated Reproduced in independent samples and at more than one site, using data not used to derive the original finding consistently associated, replicated

III: Tested experimentally

Modifying the target produced the pre-specified change in the predicted group drives, causal, modifiable target

PART II: TECHNICAL AREAS AND REQUIREMENTS

2.1 Solicited Components: SPECTRA Technical Areas

SPECTRA comprises four technical areas (TAs) that together address the five limitations of current practice, described as problem statements in Section 1.2. Proposers may submit to one or more technical areas; submissions addressing a single technical area are welcome and expected.

Proposing institutions wishing to respond to more than one TA may do so as separate submissions.

Such submissions may discuss any synergies if they exist, but each must stand on its own should only one be funded.

Across all four technical areas, SPECTRA encourages interdisciplinary approaches that combine the scientific, clinical, engineering, and lived-experience expertise each problem requires. The strongest contributions will often sit at the boundaries between disciplines. Proposers should expect to coordinate and collaborate with performers in other technical areas once the program starts.

TA Title Limitation(s) answered Function

TA1

Multi-Factor Analysis and Trajectory Identification

Limitations 1 and 2: measurement comes too late, and factors are studied in isolation

Analyze

TA2

Population-Scale Modeling and Data Integration

Limitation 3: the data that would reveal trajectories is siloed

Integrate

TA3

Precision Intervention and Trajectory Testing

Limitation 4: data does not turn into action Act

TA4

Technologies for Communication, Quality of Life, and Embedded Assessment

Limitation 5: function is measured where it least occurs

Enable

Figure 1: SPECTRA Program Structure and Technical Areas

2.1.1 Technical Area 1 (TA1): Multi-Factor Analysis and Trajectory Identification

Objective. SPECTRA addresses the first two limitations by identifying distinct autism-related trajectories and by integrating and analyzing comprehensive, multimodal datasets that resolve complex interactions among environmental, infectious, iatrogenic, genetic, epigenetic, microbiome, immune, metabolic, prenatal, and perinatal factors. These domains span biological propensity and responses to exposure and are typically measured and studied apart. Trajectory definitions may include sex, race, income, support-need level, and communication ability, with coverage reported. Thus, TA1 supplies the trajectory structure that TA3 requires in order to stratify treatment. Analyses should take a broad view that allows for identification of non-intuitive, nonlinear effects and that includes, but is not limited to, brain development as an endpoint. For example, assessments that identify stress signatures and predisposition to dysregulated immune inflammation, and connect them to developmental and behavioral trajectories, are encouraged. A core goal is individual-level trajectory assessment and causal impact of the above factors, with the understanding that initial conclusions are likely to be drawn at the subgroup level.

TA1 is an analysis technical area. TA1 will produce trajectory definitions and the interaction patterns that give rise to them; TA1 is not intended to produce clinical assessments or instruments.

TA1 is inherently interdisciplinary, requiring analytic and computational methodologists to work alongside domain scientists across genetics, exposure science, immunology, metabolism, and developmental measurement. The program may pair complementary analytic and domain strengths across awardees. Because few groups can measure the full range of relevant biomarkers on their own, proposers are encouraged to span multiple biomarker and exposure domains within a single proposal, or to team so that different combinations of markers and factors can be measured and analyzed together.

What TA1 Does and Does Not Do Three technical areas contribute to trajectory identification. The boundaries are stated here to prevent duplication.

Area Contribution to trajectories Boundary

TA1

Identifies candidate trajectories from complex interactions among biological, environmental, and developmental factors

Analyzes and applies measures. Does not develop measurement technology or assessments of any type.

TA2

Integrates and harmonizes data across institutions at population scale, and characterizes the population the data represent

Provides the data substrate, the federation, and the scale, and characterizes representativeness. Does not build individual causal or mechanistic trajectory models; that is TA1.

TA4

Develops the real-world measurement technologies that supply richer, denser, and more ecologically valid data over time

Builds and validates instruments. Does not define trajectories.

TA1 is focused on biological measurement and assessment of exposures. Performers may apply novel biological and exposure assays where existing ones are inadequate, and must justify the validity and reproducibility of any proprietary or non-standard assay methods. Development of real-world measurement technology, including home-based and device-derived capture, is the responsibility of TA4 and is not funded under TA1.

Approach: Start Immediately, Integrate Progressively TA1 begins producing results in Year 1 and does not wait for TA4 technology to mature, nor for TA2 to provide datasets. Proposers should have already identified data sets specific to their chosen approach and established partnerships or teaming arrangements with sources of retrospective data, patient cohorts for prospective studies, and biobanked samples for further analysis using methods under development in TA4.

Year 1: Analysis with existing and novel measures.

• Examine complex interactions among environmental, infectious, iatrogenic, genetic, epigenetic, microbiome, immune, metabolic, prenatal, and perinatal factors using existing data resources and established measures.

• Apply novel biological and exposure assays where existing measurement is inadequate to resolve a hypothesized interaction.

• Produce first-generation candidate trajectory definitions with explicit uncertainty, delivered to TA2 for replication at scale and to TA3 as provisional strata.

• Specify, jointly with TA4, which measurement gaps most constrain trajectory resolution, stated as a requirement on TA4 rather than a wish list.

Year 2 and beyond: Integration of TA4 measurement.

• Integrate TA4 real-world measures into the analysis as they are validated, progressively increasing the density, ecological validity, and lifespan reach of the data underlying trajectory definitions.

• Re-estimate trajectories with the enriched measurement and report how definitions change, including where earlier definitions do not survive.

• Maintain continuity of measurement so that trajectories defined in Year 1 remain comparable to those defined in Year 4.

This sequencing gives TA1 immediate traction, connects it directly to TA4, and creates the concurrent measurement period needed to establish that both new and established measures point to the same underlying causal structure.

Factors in Scope

• Genetic and epigenetic contributions, including inherited, de novo, and polygenic factors, assessed across both parents where relevant.

• Environmental, infectious, and iatrogenic exposure across development, captured prospectively where possible and integrated with host biology rather than analyzed separately.

• Microbiome, immune and inflammatory, and metabolic measurement at defined developmental windows.

• Prenatal and perinatal factors, scoped to modifiable maternal and infant health rather than to prediction of autism.

• Behavioral, developmental, communication, and functional measurement, initially from established instruments and progressively incorporating TA4-derived data.

• Social and service context, including care access, intervention exposure, and educational placement, which shape outcomes and confound naive comparison.

Analytic Requirements

• TA1 aims at defining interactions in a way that allows for inference of causality and mechanism, not merely association. Submissions that reduce to single-factor analysis across many factors are unlikely to be responsive.

• Interactions among domains conventionally analyzed in isolation should be modeled using novel tools aimed at identifying higher-order, nonlinear, and time-dependent interactions where the data support it, even if that data is not yet available, but might be produced by TA4 teams.

• TA1 submissions should aim to distinguish trajectories that arise from a single dominant driver from those arising from multiple interacting contributions.

• Characterize trajectories at the individual level with calibrated uncertainty, including within-person variability over time, in addition to clustering trajectory-based subgroups.

• Pre-register analysis plans and report candidate structures that fail, since selective reporting of surviving patterns would corrupt replication and independent validation.

• Move beyond association toward causal inference by design, using approaches such as within-family and sibling comparison, natural experiments, temporal ordering, negative controls, multisite replication, and independent validation.

Technical Area Recap

The technical areas are sequentially dependent but not sequentially executed. TA1 identifies candidate trajectories from the nonlinear interaction of many factors, including genetic, biological, behavioral, environmental, and iatrogenic (medical) influences, applying new modeling techniques to existing data rather than examining factors in isolation, and it begins in Year 1 without waiting for new measurement. TA2 builds durable, secure infrastructure that integrates and harmonizes data across institutions, characterizes the population that data represents, follows people as they age, and serves the other areas as shared infrastructure.

TA3 runs rigorous trials that act on trajectories, concentrating on the co-occurring conditions that drive the most harm, and uses each intervention as a test of the mechanism it targets.

TA4 develops the real-world measurement and support technologies that improve communication and daily function while generating the measurements TA1 and TA2 need, progressively enriching them from Year 2 onward, and it provides the delivery and endpoint-measurement technology TA3 requires.

Several feedback loops are structural (see Figure 1): TA4 returns real-world data to TA1 and TA2; TA4 returns delivery and outcome data to TA3; and, internal to TA4, assessment data improves the technologies themselves through personalization.

Deliverables

• Integration of at least three data modalities from among the factors named above.

• First-generation candidate trajectory definitions delivered in Year 1, with uncertainty bounds and pre-registered analysis plans.

• Successive trajectory revisions as TA4 measurement is integrated, with documented crosswalks establishing comparability across program years.

• A specification of the interaction patterns underlying each trajectory, in a form TA2 can test at population scale and TA3 can use to stratify.

• An annual measurement-requirements document for TA4, identifying which gaps most limit trajectory resolution.

• Analysis code and analytic protocols released for independent reproduction.

Metrics

Metric Target

Retrospective data access

By 3 months: demonstrated access to datasets sufficient to fulfil all relevant TA1 metrics, meeting requirements for balanced sex, race, income, support-need level, and communication-ability distribution.

Trajectory classification

By 6 months: at least three classification schemes, each defining at least three mutually diverging trajectories relative to a baseline of factor-matched non-autistic individuals, delivered to TA2 and TA3. By 18 months: at least one expanded (more than three trajectories) and refined scheme incorporating Year 1 marker-validation results, with at least one additional trajectory or scheme every 6 months thereafter. By 2 years:

IV&V reproduces a classification scheme from provided code and held-out data.

Interaction hypotheses

By 6 months: verifiable biomarkers or environmental factors that could predict, or serve as targets for, mechanism-based intervention for at least five non-trivial trajectories, delivered to TA2 and TA4. By 2 years: at least five refined trajectory marker sets, with at least one additional set each year thereafter.

Trajectory marker validation

By 1 year: validated association of at least three trajectory markers using held-out retrospective data, biobanked samples, prospective sampling, or a combination. By 3 years: at least two validated sets of marker interactions that, measured in combination, nonlinearly improve trajectory prediction, with one additional set in Year 4.

Measurement-gap identification

Annually: an assessment, delivered to TA4, of the measurement gaps most limiting trajectory resolution.

Out of Scope

• Analyses based on single-modality datasets, or datasets uniquely focused on a single factor.

• Single-factor association studies, or panels of single-factor analyses presented as interaction analysis.

• Development of measurement devices, home-based capture systems, or assistive technology, which fall under TA4. Teams may apply to multiple TAs.

• Construction or operation of the federated data infrastructure, which falls under TA2. Teams may apply to multiple TAs.

• Prenatal or perinatal analysis framed as autism risk prediction rather than modifiable health risk.

• Trajectory definitions derived from cohorts that systematically exclude people with high support needs or without reliable speech.

Interfaces

• To TA2: candidate trajectory definitions and interaction specifications for population-scale characterization and integration.

• To TA3: provisional strata in Year 1, refined definitions thereafter.

• To TA4: an annual statement of the measurement gaps most limiting trajectory resolution.

• From TA4: validated real-world measures for integration from Year 2 onward.

• From TA2: population-scale characterization and representativeness assessments, including how derivation cohorts differ from the broader population, against which cohort findings are judged.

2.1.2 Technical Area 2 (TA2): Population-Scale Modeling and Data Integration

Objective: Much of the data that would reveal how autism unfolds across the population are fragmented across institutions and cannot be analyzed jointly in a harmonized, privacy-preserving way. TA2 removes that barrier by serving as the data integration and federation backbone of SPECTRA. TA2 federates existing and newly generated data across institutions without moving records, applies common data standards and stringent quality control, and carries out population-and cohort-level analyses that describe prevalence, incidence and epidemiologically relevant trends across the full range of autism-affected people and support levels. Because autism unfolds over years, TA2 is built to house longitudinal, multi-domain data safely and securely over time, so that the same individuals and populations can be followed as they age. Notably, the federated data network is intended to become self-sustaining and to persist as durable national infrastructure beyond the life of the program. TA2 addresses Limitation 3, that the multi-modal data that would reveal trajectories is fragmented across institutions and cannot be analyzed together.

The federated analysis is built on an established clinical-data standard such as FHIR and OMOP and SPECTRA TA2 performers will work with the community to develop new data standards as required for new data types where established data standards do not exist. SPECTRA extends federation beyond records and claims into environmental exposure, multi-omics, behavioral, longitudinal, and sensor data and other modalities collected from TA4.

TA2 provides Shared Infrastructure:

TA2 has internal customers. Its purpose is not only to produce findings but to enable the other technical areas. A substantial share of TA2 milestones is therefore defined by whether TA1, TA3, and TA4 can execute against it.

Serves TA2 must provide Failure mode if not delivered

TA1 Harmonization of cohort measures, an analytic environment, and population-scale comparison against which cohort patterns are tested

TA1 patterns remain single-cohort findings and cannot be generalized

TA3 Trial-ready stratification, participant identification against eligibility criteria, and endpoint data drawn from routine care

Trials cannot open stratified arms and revert to unstratified enrollment

TA4 Federated learning substrate, telemetry ingestion in the common schema, and a versioned model registry

Device personalization cannot be validated, and TA4 data cannot return data to TAs 1 and 3.

External researchers

Documented access under governance, after the program's own use

The commons does not outlive the program

Submissions must state explicitly how and when these services become available.

Federation and Harmonization:

• Mapping federated data sources to a common data model. Submissions should describe their approach to mapping every federated data source into a common data model that supports federated learning across both retrospective and prospective data. The engineering effort to bring a data source into the common model is a program cost; closing the representational gap between source’s native format and the common model is part of the proposed work, not a precondition of participation.

• Extend the data model beyond its observational health care data origins. SPECTRA will collect complex, continuous, and high dimensional data types such as sensor streams, eye tracking, acoustic data, genomics, education, and environmental exposure that do not fit conventional clinical tables. TA2 will require innovative approaches to link stores with defined relationships these modalities require.

• Submissions should detail how they plan to build the privacy architecture and publish the technical architecture specifications. Specifications should include, but not be limited to, privacy-preserving linkage techniques, federated learning enablement at scale, differential privacy and secure execution. The privacy architecture must be published when operational, not merely described.

• Part of program performance will require development of a plan for financially sustainable maintenance of the infrastructure that keeps the privacy posture intact. This is expected to be developed during the program, rather than enumerated in the proposal. Proposals should detail plans for co-design of governance, consent, standards, and common data elements with the autism and data-contributing communities, so that the network reflects the priorities and privacy expectations of those whose data it holds.

Population-Scale Analysis and Methods:

SPECTRA seeks advances in method, not only application of existing methods to a new disease area. Submissions that solely apply standard supervised learning to assembled data will not be responsive.

• Population- and cohort-level statistical models that characterize prevalence, incidence and epidemiologically relevant trends across subgroups and support levels.

• Methods for measurement error, missing data and differences in ascertainment across sources, so that heterogeneous datasets can be analyzed together validly.

• Federated modeling across nodes with heterogeneous data availability, where no single site holds all modalities.

• Characterization of who is and is not represented in the federated data, and assessment of how the cohorts in which TA1 trajectories were derived differ from the broader population.

• Explicit reporting of uncertainty, bias, subgroup performance and failure, including where models do not work.

Deliverables:

• An operating federated network with a minimum of ten live nodes that has demonstrated cross-query functionality, published privacy architecture, and documented onboarding sufficient for new nodes to join without program assistance.

• Agreed common data elements, data models, and crosswalks, versioned and published under an open license for use beyond the program.

• Population characterization of the autistic population and its subgroups, including prevalence, incidence and the representativeness of the federated data, reported across the full range of support needs.

• A written report on how TA1 derivation cohorts differ from the broader population, delivered to TA1 and TA3 to bound generalizability of TA1 trajectories.

• A versioned model registry to validate TA4 personalization and to independently reproduce registered models, demonstrated by at least one independent reproduction from the registry alone.

• An analytic environment used by the other technical areas throughout the program and made available to external researchers. This environment will eventually need to be supported by a robust sustainability and transition plan that ensures documented, governed access remains available and demonstrates the long-term viability and success of the analytical environment beyond the program.

Metrics:

Nodes live Minimum ten live nodes with data flowing and cross-node query enabled by the end of program with interim targets.

Harmonization >95% of committed sources mapped to the common model and passing quality checks by end of program.

Service delivery On-time delivery of stratification, cohorts, and endpoint extracts delivered to TA3 and TA4 against dates agreed upon by the program;

target of 95% on-time, zero missed dependencies that block a TA3 trial opening.

Query performance ≤5 sec multisite query across 10+ sites.

Representativeness On-time delivery of the target autism population represented in the federated data, with gaps characterized by subgroup.

Privacy No successful re-identification under adversarial red-team testing by an independent team, repeated yearly at minimum and after each major architecture change.

Reuse 5 external research groups executing analyses against the SPECTRA platform by the end of the program.

Out of Scope:

• Centralized repositories requiring custodians to relinquish custody of their data.

• Re-aggregation of datasets already assembled by other federal programs. TA2 will consume and interoperate with those resources rather than duplicate them.

• Causal or mechanistic trajectory modeling on deeply phenotyped individuals, which is the scope of TA1. TA2 provides the sub-population and broader population context, not the individual causal model.

• Analytic environments accessible only to the performer that built them.

2.1.3 Technical Area 3 (TA3): Precision Intervention and Trajectory Testing Objective: Test interventions within trajectories rather than against the autism label, establish which options work for whom, and use each intervention as a test of the mechanism it targets. TA3 answers Limitation 4, that we do not turn data into action and treatment is chosen by trial and error.

Intervention as Causal Test

Every TA3 trial serves twice. It is a treatment trial, and it is an experimental test of a candidate driver nominated by TA1. This dual purpose is a program requirement and distinguishes TA3 from conventional intervention research.

For each arm, proposers must register in advance the target being modified, the predicted direction and magnitude of effect, the subgroup in which the effect is predicted, and the subgroup in which it is predicted not to occur. A trial that improves outcomes without testing its stated mechanism is a partial result. A trial whose prediction fails is a reportable finding and is not a program failure.

Trial Architecture

• Adaptive platform trials operating under a master protocol with a shared control, so that arms can be added and dropped without standing up new trials.

• Sequential multiple-assignment randomized designs for building treatment sequences, since real care is sequential and the operative question is what to try next when the first option does not work.

• Real-world conditions: delivery in ordinary clinics, schools and homes, broad eligibility, usual-care comparators, and outcomes measured in daily life.

• Confirmatory stages designed to support regulatory and coverage decisions.

Open Broad, Then Sharpen

Trials open with broader eligibility and add trajectory stratification as trajectories are confirmed by TA1 and their population representativeness is established by TA2. This sequencing is deliberate. It allows TA3 to begin without waiting for final trajectory definitions, and it means a delay in TA1 or the supporting data infrastructure slows the refinement of TA3 rather than halting it.

Targets and Intervention Classes

TA3 gives priority to the co-occurring conditions that cause the most harm and are most often missed, including those overlooked when their symptoms are misattributed to autism rather than investigated as distinct, treatable conditions. These include, and are not limited to, sleep disruption and disorders, seizures and epilepsy, gastrointestinal disease, chronic pain, anxiety and mood conditions, self-harm and behavioral crisis, immune and metabolic conditions, and regression or loss of function. Proposers may address other conditions where they can show comparable harm and a testable mechanism. TA3 does not prescribe which interventions to test;

proposers select them and justify each against the mechanism it targets. Because many autistic people, especially those with intellectual disability, carry a heavy and sometimes thinly evidenced medication burden, testing when to safely reduce or stop medication, not only when to start it, is within scope and is held to the same evidentiary standard.

Endpoints

Endpoints must align across the program. Each trial draws its outcome measures from the trajectory outcomes defined in TA1, expresses them wherever possible through the TA4 measures that capture behavior in daily life, and reports them so they can be read against the population characterized by TA2. Where a shared program measure exists, trials use it rather than inventing a standalone endpoint.

• Health and morbidity: e.g., seizure frequency, sleep architecture, gastrointestinal symptom burden, pain, crisis, and hospitalization.

• Function and participation: daily living, school and workplace participation, independence, and prompt dependence over time.

• Communication and language: functional communication in daily life across all modalities, including speech, augmentative and alternative communication (AAC), and gestural or behavioral signals. Measures capture initiation and not only prompted response, apply to people with and without reliable speech, and use TA4-derived real-world measures validated against structured assessment.

• Quality of life, reported by the person wherever possible and by proxy only where necessary, with the method stated.

• Family quality of life and caregiver health, measured with validated instruments rather than assumed.

• Durability at twelve months in ordinary settings. In-clinic performance at the end of an intervention period is not a sufficient endpoint.

• Medication burden and adverse drug effects, so that benefit is weighed against the harms of treatment and of polypharmacy.

• Each endpoint is defined in advance in full: the population it applies to, the outcome measured, how the analysis treats events such as rescue treatment, added medication, or dropping out, and how results are summarized across participants, so that treatment changes and attrition in real-world settings do not obscure the effect being measured.

Endpoint selection, eligibility criteria, and the definition of meaningful benefit are to be co-designed with autistic people and families, so that trials measure outcomes the community values rather than outcomes that are merely convenient to measure.

Deliverables

• An operating platform trial with at least two trajectory-stratified arms and endpoints aligned with regulatory expectations.

• At least one trajectory-matched intervention with a confirmatory result.

• Registered causal tests with results reported in full, including predictions that failed.

• A coverage evidence dossier suitable for payer review, supporting expansion of covered services.

• Trial infrastructure and protocols released for continued use after the program period.

• Targets are stated as outcomes a performer demonstrates against pre-specified thresholds fixed at award and reported by program gate. Values marked as conditions of milestone acceptance gate continuation.

Metrics

Metric Target

Stratification benefit (primary)

Pre-registered effect-size advantage for trajectory-matched treatment over unstratified care, meeting a pre-specified margin in a confirmatory arm.

Mechanism confirmation Proportion of pre-registered mechanism predictions confirmed in the predicted direction and magnitude, against a stated threshold, with disconfirmed predictions reported in full.

Stratification-rule validity The stratification rule shown to be predictive rather than prognostic and locked before the confirmatory stage, with responder-rule sensitivity and specificity reported and analytical and clinical validation demonstrated.

Sequential treatment benefit

Improvement from sequential, multiple-assignment strategies over fixed first-line care among participants who did not respond to the initial option.

Endpoint validity Trial endpoints validated against structured assessment at a pre-specified threshold using TA4-derived real-world measures, and person-reported wherever possible.

Arm throughput Time from a TA1 candidate-driver nomination to arm activation under the master protocol, against a stated threshold and reported by gate.

Enrollment representativeness

People with high support needs, intellectual disability, and without reliable speech enrolled at pre-specified minimum proportions, reported by stratum.

Safety and harm surveillance

Adverse-event and harm monitoring across all arms under an independent data safety monitoring board, with pre-specified stopping rules, and no intervention that narrows expressive range or reduces autonomy.

Durability Proportion of demonstrated benefits sustained at twelve months in ordinary settings.

Clinical significance Absolute benefit reported alongside relative effect, including number needed to treat and, where relevant, number needed to harm.

Medication burden Where medication is safely reduced or stopped, function and quality of life maintained or improved with lower medication burden and fewer adverse drug effects, against a pre-specified threshold.

Family and caregiver outcomes

Change in family quality of life and caregiver health measured with validated instruments, reported alongside participant outcomes.

Regulatory and coverage readiness

Confirmatory results structured to regulatory expectations, with coverage dossiers filed and coverage determinations obtained.

Full-result reporting All registered causal tests and predictions reported, confirmed and disconfirmed alike, with trial infrastructure and protocols released for continued use.

• Trials enrolling on the autism label alone with no stratification and no plan to add it.

• Interventions whose stated endpoint is reduction of autistic traits rather than health, function, communication, participation or quality of…