nSPECTRA AI Modeling and Federated Data Platforms

u2022 Minimum five years of experience post-doc.\n \u2022 Experience developing or evaluating machine learning and computational models with real-world health or biomedical data\n \u2022 Experience with federated or distributed learning, or with other privacy-preserving computation on sensitive data\n \u2022 Experience with longitudinal or causal modeling methods (e.g., trajectory or mixed-effects models, survival analysis, causal inference, causal ML)\n \u2022 Experience with budgeting and program or project management\n \n \nExperience/Expertise Preferred:\n \u2022 Experience building or running multi-site research data networks (e.g., OHDSI, PCORnet, N3C, or federated genomics consortia)\n \u2022 Data standards and interoperability (e.g., OMOP CDM, HL7 FHIR, GA4GH)\n \u2022 Multimodal and foundation-model approaches for combining sensor, audio, video, language and ‘omics data\n \u2022 Familiarity with health data governance and privacy requirements (HIPAA, and FERPA for school data), consent and assent for minors, and data use agreements\n \u2022 Responsible AI frameworks and regulatory context (e.g., NIST AI RMF, FDA guidance on AI-enabled medical devices)\n \u2022 Engineering for scalable platforms: cloud and high-performance computing, MLOps, security\n \u2022 Autism or neurodevelopmental data and research\n \u2022 Participatory or community data governance, i.e., giving autistic people and families a say in how their data and models are used\n \u2022 Lived experience of autism is a plus. Disclosure is voluntary and not required.\n, u2022 PhD in computer science, machine learning, statistics or biostatistics, biomedical informatics, applied mathematics, electrical engineering, or a related field. A master’s degree with substantial relevant research or industry experience may be considered.\n

ARPA-H’s SPECTRA program (Systems for Phenotypic Evaluation, Clinical Trajectories, Response, and Agency) aims to change how autism is understood and supported. It grounds research in objective, multi-factor measurement and computation instead of behavioral observation alone. By combining data across the lifespan, SPECTRA aims to match each person to the interventions most likely to help, and to keep improving care and outcomes over time (ARPA-H, n.d.-b).\n \n \nSPECTRA has four connected technical areas (TAs):\n \u2022 TA1 - Analysis: tools that link genetic, environmental, family, metabolic, immune, and developmental factors to different paths in development, health, communication, and daily function.\n \u2022 TA2 - Modeling/Integration: AI platforms that map those paths across many sites while keeping data private, and identify factors that can be changed to improve outcomes.\n \u2022 TA3 - Precision Intervention: choosing and rigorously testing interventions matched to each person.\n \u2022 TA4 - Enabling Technologies: accessible tools for communication, quality of life, and embedded assessment that measure real-world function through ordinary use.\n \n This SETA position supports TA2, the program’s integration hub. TA2 platforms will learn from biological and environmental data from TA1 and continuous real-world data from TA4 (ARPA-H, 2026). They do this by sharing models across institutions, so raw data never leaves its home site. The resulting models should separate factors that actually change outcomes from those that are only correlated. Their outputs will help TA3 choose and test interventions for each person.\n \n Specific Duties Required:\n \u2022 Provide internal subject matter expertise on federated and privacy-preserving AI for health data. Cover modeling developmental, health, communication, and functional trajectories across the lifespan. Also cover methods for finding outcome drivers that can be changed and targets that interventions can act on (ARPA-H).\n \u2022 Give insight on whether proposed approaches are conceptually and technically sound:\n \u2022 Federated architecture: federated learning and analytics across institutions and settings without moving raw data\n \u2022 Privacy: privacy-enhancing technologies, such as differential privacy, secure multiparty computation, homomorphic encryption, and trusted execution environments\n \u2022 Modeling: longitudinal and trajectory modeling, and finding subgroups of people with different trajectories\n \u2022 Causal inference: telling modifiable drivers apart from correlations, including counterfactual methods and emulating target trials\n \u2022 Multimodal combining TA1 data (genetic, ‘omics, environmental, clinical) with TA4 real-world data (sensor, audio, video, language, school-based measures)\n \u2022 Trustworthiness: validation, generalizability, bias and fairness across age, sex, race and ethnicity, and support needs, plus interpretability and uncertainty\n \u2022 Help set goals, expectations, and budgets for TA2 projects, including milestones for the platform, model performance, data onboarding, and interoperability.\n \u2022 Help review and oversee funded TA2 projects. That includes integration across the program:\n \u2022 bringing TA4 data streams into TA2 frameworks (ARPA-H)\n \u2022 bringing in TA1 data\n \u2022 providing model outputs that help TA3 choose and test interventions for each person\n \n