Job description

About Formation Bio Formation Bio is a tech and AI driven pharma company differentiated by radically more efficient drug development. Advancements in AI and drug discovery are creating more candidate drugs than the industry can progress because of the high cost and time of clinical trials. Recognizing that this development bottleneck may ultimately limit the number of new medicines that can reach patients, Formation Bio, founded in 2016 as TrialSpark Inc., has built technology platforms, processes, and capabilities to accelerate all aspects of drug development and clinical trials. Formation Bio partners, acquires, or in-licenses drugs from pharma companies, research organizations, and biotechs to develop programs past clinical proof of concept and beyond, ultimately helping to bring new medicines to patients. The company is backed by investors across pharma and tech, including a16z, Sequoia, Sanofi, Thrive Capital, John Doerr, Spark Capital, SV Angel Growth, and others. You can read more at the following links: Our Vision for AI in Pharma Our Current Drug Portfolio Our Technology & Platform At Formation Bio, our values are the driving force behind our mission to revolutionize the pharma industry. Every team and individual at the company shares these same values, and every team and individual plays a key part in our mission to bring new treatments to patients faster and more efficiently. About the Position As a Senior or Principal Data Scientist on the EHR AI team at Formation Bio, you will build the models that let us learn from real-world data (RWD): electronic health records (EHR) together with claims and other real-world sources. You will work hands-on with longitudinal EHR and claims data to predict how disease progresses, how patients respond to treatment, and which approved drugs change the course of disease. The data spans large-scale, longitudinal EHR from health systems, linked to claims, clinical notes, and imaging, alongside biobank-scale genetic and molecular data. Your results become real-world evidence (RWE) that feeds directly into which drugs we acquire, which indications we pursue, and how we design trials. This is a hands-on modeling role. You will train and evaluate models, including EHR foundation models, define the cohorts and phenotypes they depend on, and turn results into clear answers for Business Development and R&D, often on tight timelines. On the technology side, you will work closely with Engineering and Product. At the Principal level, you will also lead a workstream end to end and mentor other scientists. Responsibilities Build, train, and validate predictive models on longitudinal EHR and claims data: disease progression, time-to-event outcomes, and patient-level treatment response Benchmark and adapt EHR foundation models for patient representation against strong tabular and sequence baselines Define and validate computable phenotypes and cohorts using the OMOP CDM and standard vocabularies, including features derived from clinical notes Apply causal methods such as target trial emulation alongside predictive models to surface drugs that modify disease course Answer EHR and RWE questions from Business Development diligence and R&D programs on days-to-weeks timelines, with explicit statements of uncertainty and bias Write reproducible, reviewed code and work with Engineering and Product to turn prototypes into reusable tools Principal level: lead a modeling workstream end to end, set its technical direction, and mentor other scientists About You Required Qualifications PhD in computer science, biomedical informatics, statistics, epidemiology, or a related quantitative field (required) 2+ years of industry or research experience building machine learning models on EHR data, including linked claims or other real-world data, with substantiated results: peer-reviewed publications, deployed models, or externally validated performance Strong Python and deep learning expertise, and familiarity with data lakes and distributed compute for large clinical datasets High comfort with LLMs, including coding agents, as part of your daily work Working knowledge of how EHR and claims data are generated and structured: encounters, diagnoses, procedures, medications, labs, vitals, clinical notes, coverage periods, and reimbursement records, along with coding practices, missingness, and variation across sites, payers, and countries Fluency with clinical vocabularies and common data models, including ICD-10 and its national modifications, SNOMED CT, LOINC, RxNorm, ATC, procedure coding systems, and the OMOP CDM Familiarity with common uses of EHR data and RWE in drug development: phenotyping and cohort definition, trial feasibility and eligibility, natural history, outcome prediction, external control arms, comparative effectiveness, and safety signal detection Rigorous evaluation habits: temporal and site-held-out validation, calibration, handling of censoring, and prevention of label leakage Ability to explain technical results clearly to scientific and non-technical stakeholders Preferred Qualifications Experience pretraining, fine-tuning, or rigorously benchmarking EHR foundation models or LLM-based approaches to patient timelines Causal inference on observational health data: target trial emulation, propensity score methods, instrumental variables, doubly robust or ML-based estimators Use of LLMs for clinical text, such as phenotyping or information extraction from clinical notes Experience with OHDSI tools and network studies, or with large linked EHR and biobank resources (e.g., All of Us, UK Biobank) Experience in pharma, biotech, or health tech applying EHR models to drug development questions Exposure to health economics and outcomes research (HEOR) Total Compensation Range: Senior $185,500 - $232,000; Principal $213,500 - $267,000 Compensation Individual compensation is determined by several factors, including role scope, geographic location, and skills & experience. Your offer will reflect where you fall within the range based on these considerations. In addition to base salary, we offer equity, comprehensive benefits, and generous perks. If the posted range doesn't match your expectations, we still encourage you to apply! Where We Hire Formation Bio is prioritizing hiring in key hubs, primarily the New York City and Boston metro areas, with a hybrid model requiring 3 days per week in office. Applicants from the Research Triangle (NC) and San Francisco Bay Area may also be considered. Please apply only if you reside in these locations or are willing to relocate . Equal Opportunity Formation Bio is committed to building a diverse and inclusive team. We are an equal opportunity employer and welcome candidates from all backgrounds. All qualified applicants will receive consideration for employment without regard to race, color, creed, religion, national origin, ancestry, sex (including pregnancy, childbirth, breastfeeding, and related medical conditions), gender identity or expression, sexual orientation, age, disability, genetic information, marital status, military or veteran status, or any other characteristic protected by federal, state, or local law. We've seen a rise in recruiting impersonation scams across the industry, where fake profiles pose as recruiters aiming to trick candidates into sharing personal and financial information through illegitimate interviews and job offers. All legitimate recruiting communications from Formation Bio will come from a @formation.bio email address. If you receive outreach from someone claiming to work at Formation Bio through any other domain or channel, please disregard it and do not share any personal information.

Requirements

Must-haves

Build, train, and validate predictive models on longitudinal EHR and claims data: disease progression, time-to-event outcomes, and patient-level treatment response
Benchmark and adapt EHR foundation models for patient representation against strong tabular and sequence baselines
Define and validate computable phenotypes and cohorts using the OMOP CDM and standard vocabularies, including features derived from clinical notes
Apply causal methods such as target trial emulation alongside predictive models to surface drugs that modify disease course
Answer EHR and RWE questions from Business Development diligence and R&D programs on days-to-weeks timelines, with explicit statements of uncertainty and bias
Write reproducible, reviewed code and work with Engineering and Product to turn prototypes into reusable tools
Principal level: lead a modeling workstream end to end, set its technical direction, and mentor other scientists
PhD in computer science, biomedical informatics, statistics, epidemiology, or a related quantitative field (required)
Strong Python and deep learning expertise, and familiarity with data lakes and distributed compute for large clinical datasets
High comfort with LLMs, including coding agents, as part of your daily work
Fluency with clinical vocabularies and common data models, including ICD-10 and its national modifications, SNOMED CT, LOINC, RxNorm, ATC, procedure coding systems, and the OMOP CDM
Rigorous evaluation habits: temporal and site-held-out validation, calibration, handling of censoring, and prevention of label leakage
Ability to explain technical results clearly to scientific and non-technical stakeholders

Nice-to-haves

Experience pretraining, fine-tuning, or rigorously benchmarking EHR foundation models or LLM-based approaches to patient timelines
Causal inference on observational health data: target trial emulation, propensity score methods, instrumental variables, doubly robust or ML-based estimators
Use of LLMs for clinical text, such as phenotyping or information extraction from clinical notes
Experience with OHDSI tools and network studies, or with large linked EHR and biobank resources (e.g., All of Us, UK Biobank)
Experience in pharma, biotech, or health tech applying EHR models to drug development questions
Exposure to health economics and outcomes research (HEOR)

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aimachine learningpythonllmml

Platform: greenhouse
Effort: quick

Benefits

Benefits detected in the listing.

Vision