Principal Scientist, Translational AI Foundations

Cambridge Crossing - MA - US
Full-time
$166,770 - $202,086
Posted Nov 13, 2026

Skills

AWS
Large Language Models
Machine Learning
Plumbing
CI/CD
Communication
Python
R
Writing

Job description

At Bristol Myers Squibb, our employees often ask, “Who are you working for?”—a question that fuels collaboration, accountability, and urgency in our work. Our purpose-driven culture inspires us to discover, develop, and deliver innovative medicines to prevail over serious diseases. We offer uniquely interesting and meaningful work, opportunities for growth, and a supportive environment that values inclusion, wellbeing, flexibility, and comprehensive benefits. This is work that transforms the lives of patients, and the careers of those who do it.

When you join BMS, you are joining a high-achieving team united by a common mission.

The Informatics and Predictive Sciences (IPS) mission is to Pioneer, Partner and Predict to drive transformative insights for patient benefit. IPS conducts applied computational research in areas that include genomic, structural and molecular informatics, computational and systems biology, patient selection and translational biomarker research, and broader fields including knowledge science, epidemiology and machine learning—across the full lifecycle of drug discovery and development and across all therapeutic areas at BMS. We do this in close partnership with scientific and clinical experts in the field, both inside and outside the company.  We perform innovative science to empower key data-driven decisions across a rich pipeline of next-generation medicines. In doing so, our work transforms the lives of patients, as well as our own lives and careers.

Here, you’ll get the chance to grow and thrive through opportunities that are uncommon in scale and scope. You’ll pursue innovative ideas while advancing professionally alongside some of the brightest minds in biopharma.

Position Summary

Translational IPS is forming Translational AI Foundations (TAIF), a small team of computational biologists and AI engineers that works side by side with our translational research groups. TAIF turns the biomarker analyses scientists already perform by hand into validated, reusable AI agents and workflows, then transfers each one to the research group that will use and maintain it. The science stays with those groups; TAIF builds the shared tools that make it faster and reproducible.

We are seeking a Principal Scientist to design, build and validate the team's assay-level AI agents and the analysis logic behind them, so that a biomarker analysis that takes weeks of hands-on work today can be run, reviewed and carried to a recommendation in days. The initial focus is spatial transcriptomics and H&E, followed by other data types (ctDNA, IHC, cytokine panels, RNA-seq, PK/PD and clinical data). You will work with real program data and program scientists, and spend part of your time working directly with translational research groups. You will not be assigned to a single drug program: the output of this role is a validated, reusable tool, not a one-off analysis.

This role is for someone who has done these analyses by hand and knows where they break: a computational biologist with deep spatial and single-cell expertise who has already built LLM-based or agentic analysis workflows around that expertise, and who can show, with evidence, whether an agent's output is supported by the data or merely well-formed. You will automate established analyses with AI agents and hand them to the research groups that use them.

What you will have to work with

Spatial transcriptomics and spatial proteomics / multiplex immunofluorescence. The platforms the organization runs today: Xenium, Visium and Visium HD for spatial transcriptomics, and platforms such as COMET or PhenoCycler for spatial proteomics / multiplex immunofluorescence, backed by a pan-cancer spatial atlas license, an H&E-to-mIF platform partnership, and digital pathology and image-analysis colleagues for whole-slide H&E. These platforms are funded and running; this role exists to make their analysis reproducible and scalable.

Clinical and multimodal patient-derived datasets from BMS's early-phase clinical studies: the molecular and clinical biomarker data generated by our own early-phase trials, spanning RNA-seq, ctDNA, WES, TCR-seq and CTC, together with flow cytometry, cytokine profiling, IHC and proteomics, with the PK/PD and clinical data alongside them.

A shared engineering platform. A platform engineering scientist on the same team maintains the shared tooling: templates, automated tests, release checks, a traceable record behind every number, and the connections to governed data (Databricks), enterprise AI services and AWS cloud compute. You set the scientific standard that platform enforces; you do not have to build the plumbing.

Subject-matter experts in every assay, and designated AI contacts in each translational research group who co-develop tools with you, review them, and take them over once they are validated.

Your contributions will speed translational decisions across the BMS early clinical portfolio and help bring new medicines to patients sooner.

You will apply these resources across two areas:

•Assay agents, their expert specifications and their evaluation. The majority of the role: building the agents, writing the specification an expert signs off on, and setting the test cases and pass criteria.

•Delivery with the translational groups. The remainder: the first spatial pilots, then direct work with research groups, and maintaining each agent's validation tests after hand-off.

Key Responsibilities

Assay agents, specifications and evaluation

•Assay agents. Build and maintain AI agents for the major biomarker data types (spatial and H&E first, then ctDNA with the platform engineering scientist, IHC, cytokine, RNA-seq, PK/PD and clinical data), each with defined inputs and outputs, a traceable source for every number and a human-review step. Each agent automates an analysis a colleague already does by hand.

•Expert specifications and evaluation. Write each agent's specification with the organization's subject-matter experts: the biology being measured, data and quality control, the approved analysis flow, acceptable and unacceptable interpretations, known failure modes, gold-standard reference cases and the expert reviewer. Set the test cases and pass criteria every agent must meet before release, and run periodic peer review of agent outputs.

•Biomarker definitions. Define and validate the biomarker definitions, cutoffs and derivations for each assay, which the platform engineering scientist encodes in the team's shared ontology, so that they are applied consistently across studies; you own their scientific correctness.

•Decision-grade outputs. Own the output template: every result ships with an interpretation, its limitations and a recommendation, in a form suited to program decisions.

Delivery with the translational groups

•Pilots and expansion. Lead the initial pilots in spatial transcriptomics and H&E and measure the time saved against current practice; then extend coverage to further data types and research groups in priority order, with validation cases for every pilot.

•Research-group partnership. Work with the AI contacts in translational research groups: co-develop their first tools, review what they build for scientific correctness, and teach the methods so that others use them on their own work.

•Hand-off. Maintain each agent's validation tests after hand-off to the research group that owns it, and confirm that it still passes them.

•Scientific influence. Document methods and specifications to a standard others can build on, mentor junior team members and the AI contacts you work with, and help raise the computational and AI maturity of the broader translational organization.

Basic Qualifications

•Ph.D. in computational biology, bioinformatics, biostatistics, computer science or a related quantitative field, or in a life-science field with substantial computational training, with 4 years of relevant academic and/or industry experience

•or Master's degree with 6 years

•or Bachelor's degree with 8 years.

Preferred Qualifications

We do not expect any one candidate to bring all of the following. Depth in several of these areas, combined with the judgment to tell an output the data support from one that is merely well-formed, matters more than breadth across all of them.

•Demonstrated hands-on experience analyzing and interpreting high-dimensional molecular data from human tissue, including spatial transcriptomics (for example MERFISH, Xenium, Visium HD or CosMx) and single-cell or bulk RNA-seq, carried through to publication-ready or decision-ready results.

•Hands-on experience building LLM-based or agentic analysis workflows (tool-calling agents, multi-step LLM pipelines, retrieval-grounded annotation) and evaluating their outputs against expert review, reference results or benchmarks.

•Strong Python and/or R with reproducible analysis practice, and a record of turning recurring analyses into reusable tools that other scientists adopted.

•Clear scientific communication and experience working directly with wet-lab, translational or clinical colleagues who are not engineers.

•Hands-on analysis of patient-derived translational data in at least three of: IHC, cytokine panels, bulk or single-cell RNA-seq, ctDNA, PK/PD, spatial transcriptomics, H&E pathology; and a working understanding of pharmacodynamic, mechanism-of-action and mechanism-of-resistance questions in early clinical development.

•Spatial analysis depth: deep-learning cell segmentation, atlas-based automated cell-type annotation, spatial differential expression, neighborhood and ligand-receptor analysis; familiarity with the analysis stack (Squidpy, SpatialData, scverse) and with digital pathology tooling (HALO, QuPath).

•Evaluation design for AI output: benchmark sets and gold-standard cases, reference results, consistency checks across models, checks for claims the data do not support, and experience writing the specification an expert reviewer signs off on.

•Reproducible engineering practice: version control, tests, workflow managers (for example Nextflow or Snakemake), cloud or HPC compute (for example AWS), CI/CD, and experience packaging analyses as command-line tools, dashboards, agent tools or services that others adopted.

•Experience training and enabling scientists who are not engineers: workshops, courses, documentation, co-design, and methods that others then used on their own work.

•Prior experience in biopharmaceutical translational research or early clinical development, including biomarker strategy or clinical translational data interpretation.

•Familiarity with governed data platforms (for example Databricks) and with controlled terminologies or ontologies for biomarker data.

•A record of methods or tools evidenced by peer-reviewed publications, released software, or deployed analysis portals in spatial or single-cell biology.

We hire for skills and capabilities, not just credentials – if this role excites you, but doesn’t perfectly match your resume, we encourage you to apply anyway.

Compensation Overview:

Cambridge Crossing: $166,770 - $202,086

The starting pay range(s) listed above is for full-time employees (FTE). You may also be eligible for additional discretionary incentive cash and stock opportunities. We determine starting pay thoughtfully – carefully considering the nature of the role, required skills, work location, schedule and the knowledge and experience you bring. Final compensation is guided by pay equity principles and applicable employment laws. Compensation programs are reviewed on an ongoing basis and may be adjusted over time to reflect evolving market factors, and individual, team or Compan