Portrait of Renato Umeton

Chief of AI · St. Jude

Renato Umeton, Ph.D.

Vice President of Data Sciences and Chief of Artificial Intelligence, Office of Data Science, St. Jude Children's Research Hospital

We are hiring

Building the AI and data science team at St. Jude

St. Jude has more than 20 open data science and AI roles in Memphis (St. Jude careers, October 2026), including an Agentic AI Scientist and roles in survivorship data science (faculty), bioimage informatics, genomics infrastructure, AI product management and quantum-focused education. Open roles and the fastest route to me are on LinkedIn.

See open roles and get in touch on LinkedIn → All St. Jude jobs →

Renato Umeton is Vice President of Data Sciences and Chief of Artificial Intelligence in the Office of Data Science at St. Jude Children's Research Hospital, which he joined in 2025 as its first Chief of AI.

Before St. Jude he was Director of AI Operations and Data Science Services in the Informatics & Analytics department at Dana-Farber Cancer Institute, a teaching affiliate of Harvard Medical School, where he built the department's AI and data science practice from one person to roughly forty people at its peak. He holds academic appointments at the Harvard T.H. Chan School of Public Health and MIT.

The record below is newest first, and read in that order it runs backward through a career. The most recent papers put large language models in front of oncologists and pathologists; the earliest, from 2007, model lava flows with cellular automata on the slopes of Mount Etna. The turn from physics toward medicine came from something Renato recognized early: he had excellent technical answers but, on his own, poor questions — and the best questions belong to biologists and clinicians. Working alongside them is what gave the methods somewhere consequential to land. The constant underneath is optimization and machine learning aimed at systems too large, too noisy, or too consequential to reason about by hand — physical systems first, then biological ones, and now the clinical and operational systems that surround a patient.

He earned a B.S. and an M.S. in Computer Science and a Ph.D. in Mathematics and Informatics, all at the University of Calabria, Italy.

  • 165 works
  • 2007–2026 (20 years)
  • 72 venues
  • 141 linked to publisher
20072026
One segment per year, oldest at the left, each segment as wide as that year's output. Jump to a year:

Publications

2026

6 works

2025

14 works

2024

7 works

2023

10 works

2022

17 works

2021

16 works

2020

11 works

2019

16 works

2018

11 works

2017

6 works

2016

3 works

2015

3 works

2014

4 works

2013

13 works

2012

11 works

2011

7 works

2010

5 works

2008

4 works

2007

1 work

Published in

72 journals, conferences and preprint servers carry this work, from clinical oncology and neurology to bioinformatics, systems biology and machine learning.

  • Multiple Sclerosis Journal10
  • Cancer Research8
  • bioRxiv8
  • Journal of Clinical Oncology5
  • Clinical Cancer Research5
  • Journal of Thoracic Oncology5
  • arXiv4
  • Annals of Oncology4
  • MICCAI3
  • Laboratory Investigation3
  • Journal for Immunotherapy of Cancer3
  • JCO Precision Oncology3
  • Journal of Neuroimmunology3
  • Neurology3
  • PLoS One3
  • Nature Medicine2
  • Scientific Reports2
  • Molecular Cancer Research2
  • Frontiers in Neurology2
  • ACS Synthetic Biology2
  • International Conference on Artificial Immune Systems (ICARIS)2
  • European Modeling and Simulation Symposium (EMSS)2
  • Italian Workshop on Artificial Life and Evolutionary Computation (WIVACE)2
  • Nature Communications1
  • Neuro-Oncology1
  • PLOS Digital Health1
  • Proceedings of Machine Learning Research1
  • Proceedings of the National Academy of Sciences1
  • Heliyon1
  • The Lancet Digital Health1
  • NEJM AI1
  • Nature Machine Intelligence1
  • Communications Engineering1
  • Journal of Pain and Symptom Management1
  • BJUI Compass1
  • Cell Reports Medicine1
  • Bio-IT World1
  • International Journal of Molecular Sciences1
  • European Journal of Neurology1
  • JAMA Oncology1
  • JMIR Medical Informatics1
  • JCO Clinical Cancer Informatics1
  • Informatics1
  • NeurIPS1
  • Journal of the Neurological Sciences1
  • Journal of Medical Internet Research1
  • Artificial Intelligence in Medicine1
  • Genes1
  • American Medical Informatics Association (AMIA)1
  • Mastology1
  • Neurotherapeutics1
  • Oncotarget1
  • Stroke1
  • Cytokine & Growth Factor Reviews1
  • BioMed Research International1
  • Annals of Clinical and Translational Neurology1
  • Biophysical Journal1
  • International Journal of Genomics1
  • BMC Bioinformatics1
  • Human Microbiome International Conference1
  • International Workshop on Bio-Design Automation (IWBDA)1
  • Advances in Systems Biology1
  • Design Automation Conference (DAC)1
  • Proceedings of the1
  • Bioinformatics Italian Society (BITS)1
  • Workshop on Process Algebra and Stochastically Timed Activities (PASTA)1
  • IEEE Transactions on Biomedical Engineering1
  • IEEE International Conference on BioInformatics and BioEngineering (BIBE)1
  • World Congress of Biomechanics (WCB)1
  • Journal of Biotechnology1
  • International Workshop on Ontology Repositories and Editors for the Semantic Web (ORES)1
  • Geophysical Research Abstracts1

The role at St. Jude

St. Jude treats data science and AI as a strategic priority. Its 2022–27 Strategic Plan sets, as Strategic Initiative 3, the goal to “become a leader in the application of data science to biological discovery”. In October 2024 the hospital named its first Chief Data Scientist to lead the newly formed Office of Data Science, which fulfills that goal: a $195 million research enterprise that the hospital's news release said would add 115 positions. In 2025 Renato joined it as St. Jude's first Chief of AI, reporting to the Chief Data Scientist and Senior Vice President of Data Science. He is based in Greater Boston and is hiring in Memphis.

Four men talking in a St. Jude auditorium, in front of a large screen showing the KIDS25 logo of the Office of Data Science.
At KIDS25, the Office of Data Science's Knowledge in Data Science symposium: Drs. Downing, Umeton and Babu, with Dr. Natarajan of Google DeepMind.

The hospital has used these methods for a long time: its researchers applied advanced image analysis to immune cell studies in 1990, and by the early 2000s machine learning was classifying subtypes of acute lymphoblastic leukemia and linking them to patient outcomes.

His remit covers four areas

  • Governance

    Governance for every data set, platform and service, whether it belongs to one lab, a department or the whole hospital, so each one is found and used well. Unified data governance bridges fundamental science, translational research, clinical practice and global efforts.

  • Data catalog

    One place to find St. Jude's data: repositories unified, created and kept current, with navigators who know the data and guide others to it.

  • Platforms

    Software for research, operations and clinical work, bought or built. Azure Databricks is one starting point among several.

  • Services

    Training, advice, consulting and hands-on support so teams can bring AI into daily work, with data scientists who can sit inside departments and work next to biologists.

His remit covers four areas: governance, data catalog, platforms, and services.

Around those four sit workforce education and capacity building, data stewardship, and responsible, secure AI. Safeguarding patient and laboratory data stays at the forefront while innovation moves ahead. St. Jude has spent decades reducing the risks of innovation while maximizing its benefits, with a science-driven, iterative approach, and it follows a generative AI policy that matches each tool to its intended use and keeps high-risk clinical uses apart from research and productivity tasks. The long-term aim is large, linked data collections that make AI work possible. The pieces are connected, and the work needs collaboration across the St. Jude community, guided by a long-term vision that puts patients first.

More in St. Jude's interview, “5 Questions for Renato Umeton, PhD” (May 2025). Or jump to the full list of publications.

Selected work

Six threads run through Renato's publication record. Each lists a few representative papers; the number after each one opens its full entry on the Scientific works tab.

Software and platforms

Software from Renato's teams, collaborations and papers, with his part in each stated plainly. Most are open source. Paper numbers open the full entry on the Scientific works tab.

  • The MyGPT workspace with labelled callouts: a chat answer with relevance and hallucination scores and its sources on the left, the document library in the middle, and the cited PDF page with highlighted passages on the right.
    Image: MyGPT project, GPL-3.0

    MyGPT 2026

    An open-source St. Jude platform that runs open language models locally, cites its sources and scores each answer for relevance and hallucination. Co-author.

    codesitepaper #165
  • Four-panel cartoon: an AI agent alone picks answer A; two confident peer agents assert B; the agent follows the peers and picks B; a referee re-asks the agent alone, gets A, and raises a flag because the answers differ.
    Image: benchmaxxing project, MIT License

    Agents Catching Agents 2026

    Code and data showing how clinical AI agents pass shortcuts to each other and game benchmarks, with a referee that re-asks each question without the pressure and flags answers that change. Co-author and contributor.

    codesitepaper #164
  • The croissant-baker documentation home page, with its croissant-in-a-chef's-hat logo and a one-paragraph description of the tool.
    Image: croissant-baker project, MIT License

    Croissant Baker 2026

    Writes Croissant machine learning metadata straight from a folder of files, including DICOM, NIfTI, FHIR and genomics formats. Code contributor (MIT-LCP).

    codesite
  • The M3 system architecture: a client sends natural-language queries to the M3 server, whose layers handle the protocol, security and validation, and data access, with core and clinical tools, over a local SQLite copy or Google BigQuery holding MIMIC-IV.
    Image: M3 project, MIT License

    M3 and M4 2025–2026

    Lets AI assistants query the MIMIC-IV clinical database in plain language through the Model Context Protocol, locally or in the cloud. Co-author and code contributor.

    M3 codeM4 codepaper #162
  • The GPT4DFCI web app in a browser: a list of past conversations on the left and the Dana-Farber welcome panel with the rules of use on the right.
    Image: GPT4DFCI project, GPL-2.0

    GPT4DFCI 2023–2025

    A private, secure generative AI assistant inside Dana-Farber's network, used for operations, administration and research and kept apart from direct patient care. He was product owner, is first author of the paper and published the code.

    codeAPInewspaper #143
  • World map of federated evaluation: one AI model is sent to hospitals on several continents, each reports its own accuracy, and the sites where accuracy drops are circled in red.
    Image: MedPerf project, Apache-2.0

    MedPerf 2021–

    Open benchmarking for medical AI: models are tested on data held at each hospital, and the data never leave the site. Built by the MLCommons Medical AI working group he co-chairs; co-first author of the Nature Machine Intelligence paper and product owner.

    codedocspaper #138paper #149
  • Hexagon diagram of GaNDLF's features around a central GaNDLF hexagon: input/output and anonymization, harmonization, augmentation, network topologies, training and post-processing, each with radiology and histology options.
    Image: GaNDLF project, Apache-2.0

    GaNDLF 2021–

    A PyTorch framework for segmentation, regression and classification on medical images with little or no code, now part of MLCommons. Co-author.

    codedocspaper #136
  • PathML overview diagram: slides in many imaging modalities and file formats are loaded as SlideData, cut into tiles and preprocessed in parallel by a pipeline of transforms, saved as .h5path files and fed to a deep learning model through a DataLoader.
    Image: PathML project, GPL-2.0

    PathML 2019–

    An open-source Python toolkit for computational pathology: whole-slide images, multiplex image analysis and machine learning pipelines. Developed with the Loda lab and hosted by his former Dana-Farber group; co-author and product owner. It ships a Digital Pathology Assistant that helps researchers write PathML analyses.

    sitecodepaper #116paper #153
  • OREMP running inside the Cytosolve web interface at its 'Align Models' step, listing reactions that two BioModels pathway models share.
    Image: Umeton, Nicosia and Dewey, BMC Bioinformatics 2012, Figure 4, CC BY 2.0

    OREMP and OREMPdb 2010–2012

    Merged computational pathway models with biomedical ontologies; run over BioModels it produced OREMPdb, a dictionary of 7,360 species sets. First author, at MIT.

    workshop paperpaper #28

  • dlvhex RDF plugin 2005

    Lets HEX logic programs in the dlvhex solver read RDF data from the semantic web. He wrote the first version of its @RDF builtin as a visiting scholar at TU Wien.

    code

Personal projects, all on GitHub:

  • One simulated patient over ten years: a particle in a double-well curve with wells labelled Health and Flare, the weekly record of flares, and below them an illustrative disability score over the same years.
    Image: msrelapse project, MIT License

    msrelapse: a Python version of a 2013 multiple sclerosis model (paper #33)

  • Infographic titled Bio-Croissant: Standardizing Metadata for Biomedical AI, with a bridge and a croissant drawn as a biomedical map, and panels on OMOP CDM support and on bioimaging and pathology metadata.
    Image: Bio-Croissant to OMOP project, MIT License

    Bio-Croissant to OMOP: a prototype metadata format for biomedical data sets, with a converter to the OMOP data model

  • The ztop terminal window split into panes: two process monitors on the left, one sorted by CPU and one by memory, a Mac metrics panel with CPU, GPU, memory and power at top right, and a container list at bottom right.
    Image: ztop project, MIT License

    ztop: a terminal system monitor for macOS

  • Two views of the aitampermonkey menu: on the left its list of actions, such as summarize, save as Markdown and send to a chatbot; on the right the submenu for a basket of pages. A round icon below the menu opens it.
    Image: aitampermonkey project, BSD-3-Clause

    aitampermonkey: a browser userscript that adds a floating menu to every web page, to send the page or a selection to AI chatbots, summarize it, or save it as Markdown

  • The SafeClaw home page on a dark background, with the headline Browser-Native Personal AI Assistant, a short description and buttons to launch the app or view the code.
    Image: SafeClaw project, MIT License

    SafeClaw: an AI assistant that runs entirely in the browser, with cloud or local models (alpha)

  • The Fluid demo app at its Customize view: a change requested in plain language, the steps of the run, a summary of the changed files and the tests it suggests before the change can go live.
    Image: Fluid project, MIT License

    Fluid: a prototype where every user gets their own version of an app and changes it in plain language; a change goes live only if the app's tests pass

Career and service

Renato's role at St. Jude and his other current roles, newest first, then his career before St. Jude and his education.

Current

Career before St. Jude

  • 2018–2025
    Dana-Farber Cancer Institute: Manager, Data Science (2018–2021), then Associate Director (2021–2023) and Director (2023–2025) of AI Operations and Data Science ServicesIn Boston. Founded the group in the Informatics & Analytics department, reporting to the Chief Data and Analytics Officer, and grew it from one person to about 40 at its peak, temporary staff included. Long term, it bridged research and the clinic with AI built and deployed for the institute; short to medium term, it gave faculty tailored AI and data science support in their labs, centers and departments. From interns to area leads, the team brought AI, data engineering, abstraction and labeling, software, cloud operations, and medical and nursing practice. Its work: automated, multimodal, self-supervised and federated machine learning; large language models and NLP; computer vision on pathology, MRI and CT images; MLOps across clouds; cloud and AI strategy; client management; and tools that take researchers from “X” to “AI-powered X”, among them the institute-wide GPT-4 deployment (GPT4DFCI).
  • 2020–2025
    Affiliate, Weill Cornell MedicineHis teams supported the Department of Pathology and its research divisions in digital and computational pathology, cloud and hybrid research infrastructure strategy, education and data science.
  • 2015–2024
    Bioinformatics and data science consulting
  • Jun 2016–Jun 2018
    Bioinformatics Data Scientist, Mass General BrighamProjects with teams across Partners HealthCare (now Mass General Brigham), Brigham and Women's Hospital and Dana-Farber, with collaborators in clinical and translational informatics, medical oncology, oncologic pathology, cancer biology, cancer immunology and virology, psychosocial oncology and palliative care, and pharmacy. Some were later packaged into institutional research offerings such as the Tumor Mutational Burden data product and the OncDRS Connector; others were considered to improve clinical operations in bioinformatics sequencing and in pharmacy operations and cost optimization.
  • 2013–2015
    Director of Bioinformatics, CytoSolveIn Greater Boston.
  • 2011–2013
    Postdoctoral researcher, Sapienza University of Rome and Sant'Andrea hospitalIn Rome, Italy: clinical and translational research, including multiple sclerosis genetics.
  • 2010
    Ph.D., University of Calabria; visiting scholar at MIT and Microsoft ResearchMIT Biological Engineering (Apr 2009–Oct 2010) and Mechanical Engineering (2009–2010), in Greater Boston, and the Microsoft Research – University of Trento Centre for Computational and Systems Biology (2009), where he modeled the Toll-like receptor, a complex part of the human immune system, with parallel distributed simulations.
  • 2005–2008
    Teaching Assistant, University of Calabria
  • 2005
    Visiting scholar, TU Wien; intern, Lixto, both in Vienna, AustriaIn TU Wien's Knowledge Representation group he wrote the first version of the @RDF builtin for dlvhex, which let DLV logic programs import RDF ontologies from the semantic web (see the Software tab). At Lixto he worked on semantic data extraction from the web with TU Wien's Database and Artificial Intelligence group; Lixto was later acquired by McKinsey & Company and became part of its Periscope offering.

Education

Service, honors and certifications

  • Invited speaker for US federal government departments and Fortune 20 companies.
  • Peer reviewer for journals of NEJM Group, Nature Portfolio, Cell Press, IEEE, ACM and Oxford University Press.
  • Mentor to more than 50 people, from bootcamp students to MD-PhD candidates, including data science and data engineering mentees in Correlation One's “Data Science for All Women” program (Jan 2019–Dec 2022), which aims to open AI and data science jobs to more people; the work also supported his group's aim of hiring skilled professionals from diverse backgrounds.
  • About 19 years in AI, data science and big data (as of 2026), in academic medical centers, consulting and industry, as scientist, manager and executive. Co-inventor on six patent applications, listed on the Scientific works tab, and on many invention disclosures; some of his work has been used in clinical trials or licensed.
  • Takes part in academic, industry and government collaborations that aim to widen the good AI and machine learning can do in healthcare.
  • MIT-Italy Fellow, part of MISTI (MIT International Science and Technology Initiatives), after winning the yearly competition of the Italian Mathematics Society.
  • Engineer of National Interest and Individual with Extraordinary Ability, as recognized by the United States Department of State and Department of Homeland Security.
  • Certifications: DS4A Empowerment Mentor Badge, Data Science for All (DS4A) Mentor and DS4A Data Engineering Mentor (Correlation One); IBE Entrepreneurship Program; Mastering the Leadership Challenge.
  • NEJM Group
  • Nature Portfolio
  • Cell Press
  • IEEE
  • ACM
  • Oxford University Press
  • Correlation One
  • MISTI
  • Italian Mathematics Society
  • Department of State
  • Department of Homeland Security

News

Press coverage and interviews that name Renato Umeton, newest first.