CV
Jonathan L. Golob, MD, PhD

Jonathan L. Golob, MD, PhD

Physician-scientist and software engineer

With a view from the bedside, the bench and the codebase, I find what’s missing, build the fix, and leave it working. I lead applied AI/ML for vaccines and infectious disease at GSK, keep a limited clinical practice in transplant infectious disease, and still write the software myself.

Now: Director of AI/ML for Vaccines & Infectious Disease at GSK; leads and is accountable for GSK's AI/ML and data engineering across all six Fleming Initiative Grand Challenges; GSK scientific co-lead for Antibiotics and Antifungals.

Interested in ambitious collaborations at the intersection of frontier AI, biomedicine, and translational R&D.

What I bring

AI and evaluation

  • ViroGym (NeurIPS 2026 Spotlight), ChoreCode, the preterm-birth DREAM Challenge, and two first-author benchmarking papers in BMC Bioinformatics.
  • The National Science Foundation (NSF) selected the Microbiome Preterm Birth DREAM Challenge dataset, developed and co-released with MaLiAmPi, as one of its core datasets for the National Artificial Intelligence Research Resource (NAIRR) Pilot program.
  • Personally develops and directs work with protein and RNA language models, diffusion approaches, and uncertainty-aware biological-sequence modeling.
  • Forty-plus public repositories, four packages on PyPI, one workflow maintained for eight years.

Biology and clinical medicine

  • A physician at the intersections of medicine: infection, oncology, transplantation and immunology. Board-certified in internal medicine and infectious disease, with a limited active practice caring for patients with cancer.
  • Research across those boundaries: graft-versus-host disease, cancer immunotherapy, inflammatory bowel disease, preterm birth and critical care.
  • Clinical trials from protocol and biomarker design through site PI and global unblinded safety physician.
  • Bench training in stem-cell epigenetics, gene-therapy vector design and organoid models; 78 publications, including preprints and abstracts, and more than 4,100 citations (Google Scholar, November 2026).

Leading people and programs

  • Leads cross-functional teams of 20–30 by persuasion rather than line authority, alongside a direct team of three to five of mixed seniority, where priorities move quickly.
  • Develops people. Junior engineers on my team have become productive within about a year, in a group where 18–24 months is typical; a member of my team was first author on a NeurIPS Spotlight paper.
  • Regularly brought in as a reviewer or working scientist on executive-level questions.
  • Builds teams in which people and AI systems complement each other.

How I work

At the confluence of the bedside, the bench and the codebase, I make new things possible.

Making microbiome studies comparable

MaLiAmPi → harmonized DREAM Challenge dataset → NSF NAIRR Pilot selection
The problem
Microbiome studies could not be combined. Each used different laboratory and sequencing methods, so a model trained on one study failed on the next.
What was missing
A stable, shared set of features that did not depend on taxonomy or on how the samples were processed.
The fix
Designed and built MaLiAmPi, an open-source workflow that turns raw data from technically different studies into one feature set.
What it enabled
The harmonized, multi-study dataset behind the Microbiome Preterm Birth DREAM Challenge, with models validated on independent data. In 2025 NSF selected that dataset, one of 10 chosen through a competitive process with 12 federal agencies, for integration into the National Artificial Intelligence Research Resource (NAIRR) Pilot. It also supported the Stabl biomarker-discovery method (Nature Biotechnology), and the method is the subject of a pending patent.
My role
Senior author and lead developer of MaLiAmPi; first author and co-lead of the DREAM Challenge; co-author on Stabl.

From a cancer ward observation to clinical trials

Microbiome and stem-cell transplant · Joel Meyers Endowment Scholar
The problem
Graft-versus-host disease is a leading cause of harm after stem-cell transplant for blood cancers, and there was no early way to tell who would develop it.
What was missing
A measurable signal before the disease appears, and a mechanism that could be acted on.
The fix
Showed that the gut microbiome at the time of blood-count recovery predicts severe graft-versus-host disease, then combined patient data with organoid models to tie bacterial butyrate to impaired gut repair.
What it enabled
Dietary intervention trials designed to change the microbiome in stem-cell transplant (Nature Medicine) and alongside immunotherapy for melanoma.
My role
First author on both studies; biomarker lead and protocol co-author on the trials.

A benchmark for viral proteins

ViroGym · NeurIPS 2026 Spotlight
The problem
Protein language models were being proposed for vaccine decisions, but the standard benchmarks contain few viral proteins.
What was missing
A test of whether performance in the laboratory predicts what happens as viruses evolve in the real world.
The fix
Conceived the benchmark and handed the idea to a member of my team, who led the work as first author.
What it enabled
A NeurIPS 2026 Spotlight paper and a public benchmark for choosing models for vaccine work.
My role
Originated the project; supervised the first author; second author.

Evidence and infrastructure in the first months of COVID-19

2020 · University of Michigan · ISARIC
The problem
Critically ill patients were arriving before there were tests, treatments or shared data.
What was missing
A validated antibody test, early evidence on treatment, and a common international dataset.
The fix
An early, critical contributor to the ISARIC COVID-19 dataset, and co-founder of the University of Michigan COVID-19 case series. Designed and validated SARS-CoV-2 serology for the clinical laboratory. Co-authored an early study of tocilizumab in ventilated patients, while providing frontline critical care.
What it enabled
A global standardized dataset used across many later studies; a working clinical test; and early evidence for IL-6 blockade, followed by site-PI roles on treatment and vaccine trials.
My role
Clinician, investigator and builder.

Models that inform vaccine development decisions

GSK · 2024 – present
The problem
Vaccine development and lifecycle decisions depend on how a virus will evolve and when infections will occur.
What was missing
Predictive models that decision-makers could use, with the uncertainty stated.
The fix
Built deep-learning models predicting viral evolution, infection incidence and case-accrual timing, and host–immune interactions, and led their transition into maintained tools.
What it enabled
Used to guide development and lifecycle decisions for a $1B vaccine program; related approaches supported additional programs in the $100M–$500M range.
My role
Technical lead and manager of the ML engineering team.

Evaluating AI coding models on real maintenance work

ChoreCode · independent project
The problem
Leaderboards for coding models said little about routine maintenance work, what it costs, or how often a model fails without saying so.
What was missing
An evaluation with hidden tests, checks on the scorer itself, tasks that cannot be solved, and cost counted alongside correctness.
The fix
Conceived, designed and built ChoreCode end to end.
What it enabled
Public, downloadable results across many model configurations and an API for choosing a model by quality and cost.
My role
Sole designer and developer.

Experience

Director, AI/ML — Vaccines & Infectious Disease
GSK · R&D · AI/ML

I lead applied AI/ML and data engineering within the Biomedical AI group in R&D at GSK. Joined as Lead Engineer, AI/ML in June 2024; promoted to Director in March 2025.

  • Models that inform decisions — built deep-learning models predicting viral evolution, infection incidence and case-accrual timing, and host–immune interactions, used to guide development and lifecycle decisions for a $1B vaccine program; related approaches supported additional programs in the $100M–$500M range. Leads their transition into maintained tools for clinical-development decision-makers.
  • Fleming Initiative — leads GSK’s AI/ML and data engineering across all six Grand Challenges; GSK scientific co-lead for Antibiotics and Antifungals, co-developing their scopes of work and scientific plans.
  • Teams — built and directly manages a specialist ML engineering team, and repeatedly forms and directs cross-disciplinary teams of 20–30.
  • Technical judgment — advises senior R&D leaders on problem framing, method selection, evaluation and build–buy–partner decisions; personally develops and directs work with protein and RNA language models, diffusion approaches, and uncertainty-aware biological-sequence modeling.
  • Responsible AI — piloted a project for GSK’s Responsible AI programme and authored a perspective on the work.
Assistant Professor
University of Michigan · Internal Medicine / Infectious Diseases

I led a research group focused on the practical clinical translation of microbiome science to patients within precision medicine — pairing novel computational techniques with cutting-edge in vitro and in vivo models — and responded to the COVID-19 pandemic, caring for hundreds of critically ill patients while developing and running human observational and interventional trials.

Senior Fellow · Joel Meyers Endowment Scholar
Fred Hutchinson Cancer Center / University of Washington

Postdoctoral fellow studying how the gut microbiome relates to and mechanistically contributes to outcomes in patients undergoing hematopoietic stem cell transplant. Mentor: David Fredricks.

  • Combined human observational study data with novel analytic and computational techniques.
  • Attending physician specializing in cancer-related infections.
  • Retained an affiliate appointment at Fred Hutch, 2018 – 2022, while on the faculty at Michigan.
Doctoral Researcher (MD–PhD)
University of Washington · Department of Pathology

My doctoral work used the (then-new) pluripotent stem-cell model to establish the basic epigenetic mechanisms behind the earliest stages of human development. I employed some of the earliest “next-generation” high-throughput sequencing and built a novel computational pipeline integrating transcriptional micro-array and tiling-array data. Mentor: Charles “Chuck” Murry.

Undergraduate Researcher
Johns Hopkins University · School of Medicine

Worked in research laboratories from my first year, alongside dual degrees in biomedical engineering and computer science. I began in Paul Kessler’s laboratory on adenoviral and adeno-associated virus (AAV) vectors, then joined Linzhao Cheng’s laboratory, where by my third and fourth years I was helping to develop some of the early lentiviral vectors for gene transfer into human blood stem cells.

Selected publications

Highlights from 78 papers across 9 domains · full list on ORCID & Google Scholar.

  1. Y. Zhou, J. L. Golob, A. M. Karimi Mamaghan, S. Bauer, and P. Schwab, “ViroGym: Realistic Large-Scale Benchmarks for Evaluating Viral Proteins,” NeurIPS 2026 (Spotlight), doi: 10.48550/arXiv.2603.06740
    Author 2 of 5NeurIPS 2026 · SpotlightOriginated the project and supervised the first author through to a Spotlight.
  2. J. Hédou et al., “Discovery of sparse, reliable omic biomarkers with Stabl,” Nat Biotechnol, Jan. 2024, doi: 10.1038/s41587-023-02033-x
    Author 18 of 28
  3. J. L. Golob et al., “Microbiome preterm birth DREAM challenge: Crowdsourcing machine learning approaches to advance preterm birth research,” Cell Rep Med, p. 101350, Dec. 2023, doi: 10.1016/j.xcrm.2023.101350
    First author
  4. Mary Mansour Riwes, Jonathan L. Golob, John Martin Magenau, … Pavan R. Reddy. Feasibility of a dietary intervention to modify gut microbial metabolism in patients with hematopoietic stem cell transplantation, Nature Medicine, 2023. doi: 10.1038/s41591-023-02587-y
    Author 2 of 19
  5. S. S. Minot et al., “MaLiAmPi enables generalizable and taxonomy-independent microbiome features from technically diverse 16S-based microbiome studies,” Cell Reports Methods, p. 100639, Nov. 2023, doi: 10.1016/j.crmeth.2023.100639
    Senior author
  6. J. L. Golob and S. S. Minot, “In silico benchmarking of metagenomic tools for coding sequence detection reveals the limits of sensitivity and precision,” BMC Bioinformatics, vol. 21, no. 1, p. 459, Oct. 2020, doi: 10.1186/s12859-020-03802-0
    First author
  7. E. C. Somers et al., “Tocilizumab for treatment of mechanically ventilated patients with COVID-19,” Clinical Infectious Diseases, p. ciaa954, Jul. 2020, doi: 10.1093/cid/ciaa954
    Author 4 of 22
  8. J. L. Golob et al., “Butyrogenic bacteria after acute graft-versus-host disease (GVHD) are associated with the development of steroid-refractory GVHD,” Blood Adv, vol. 3, no. 19, pp. 2866–2869, Oct. 2019, doi: 10.1182/bloodadvances.2019000362
    First author
  9. J. L. Golob et al., “Stool Microbiota at Neutrophil Recovery Is Predictive for Severe Acute Graft vs Host Disease After Hematopoietic Cell Transplantation,” Clin. Infect. Dis., vol. 65, no. 12, pp. 1984–1991, Nov. 2017, doi: 10.1093/cid/cix699
    First author
  10. J. L. Golob, E. Margolis, N. G. Hoffman, and D. N. Fredricks, “Evaluating the accuracy of amplicon-based microbiome computational pipelines on simulated human gut microbial communities,” BMC Bioinformatics, vol. 18, no. 1, p. 283, May 2017, doi: 10.1186/s12859-017-1690-0
    First author

View the complete publication list in the long-form CV →

Clinical medicine, trials & pandemic response

Clinical expertise

A clinician trained at the deepest end of infectious disease — caring for the most immunocompromised patients (cancer, solid-organ and stem-cell transplant), where an infection is often the difference between life and death.
  • Transplant infectious disease — solid-organ & hematopoietic stem cell transplant
  • Care of immunocompromised hosts
  • Oncology & cancer-related infections
  • General infectious disease & internal medicine
Clinical Assistant Professor — limited active practice caring for patients with cancer-related infections
University of Washington · Current
Attending physician — cancer-related infections
Fred Hutchinson Cancer Center · 2016 – 2018
Attending physician (Consultant of the Month, 2015)
Harborview Medical Center · 2013 – 2018
Board certified:ABIM — Internal Medicine (2014)ABIM — Infectious Disease (2016)

Clinical trials

Prebiotic microbiome intervention with immune-checkpoint inhibitors (melanoma)
Protocol co-author / biomarker-strategy lead · Investigator-initiated Phase 1 — completed
Prebiotic microbiome intervention in hematopoietic stem cell transplant
Biomarker analysis lead · Phase 0/1
Mesenchymal stromal cells for moderate-to-severe COVID-19 ARDS
Site PI / organizing committee · Randomized trial
COVID-19 vaccine trials
Site PI · Multiple trials
Global vaccine-development programs
Global unblinded safety physician · Multiple trials

Pandemic response

Among the first to care for critically ill COVID-19 patients at the University of Michigan, and for hundreds over the course of the pandemic — while simultaneously standing up the clinical research and laboratory infrastructure to fight the pandemic.
  • Authored an early case series identifying IL-6 blockade (tocilizumab) as a viable therapy for severe disease.
  • Designed and validated SARS-CoV-2 serology testing for the University of Michigan clinical laboratory.
  • The COVID-19 trial roles from this period are listed under Clinical trials above.

Safety, evaluation & responsible AI

Responsible AI work spanning empirical model evaluation, biological risk, and governance in high-stakes scientific systems.

AI evaluation in practice

Conceived, designed, and built an independent AI evaluation research and engineering project using realistic tasks, hidden tests, anti-gaming constraints, structured failure analysis, cost-aware comparison, and API-driven recommendations.

Deep learning for biodefense

Develops deep learning models to support biodefense and detect anomalous or potentially concerning uses of AI in health and biological contexts.

Responsible AI practice

Applies responsible AI practice to clinical AI safety, model evaluation, and governance for tools deployed in high-stakes scientific and healthcare settings.

Writing, talks & media

Writing & science communication

Two decades translating science for the public.

Delayed Critical featured

Researched creative work on nuclear safety, institutional accountability, and the failure of technical organizations to act on known risk — themes that echo my work in AI governance and sociotechnical risk.

The Stranger

Answered readers' science questions; extensive coverage of the 2008 financial crisis and the 2011 Fukushima disaster.

Ars Technica

Long-form science & technology features for a general audience.

The Conversation

Public-facing analysis from a physician-scientist.

Medium — LLMed AI

Decoding the genetic jigsaw: interpreting variants of unknown significance and cancer screening.

Talks & media 22

Fleming Initiative Science Summit on fungal disease and antifungal resistance, Royal Institution, London · Jun 2026
The Nobel Prize to now: 80 years since the penicillin Nobel — Fleming Initiative, Ineos Oxford Institute and GSK, Royal Society of Medicine, London · Nov 2025
The Evolution and Spread of Antimicrobial Resistance Alleles
NIH / BV-BRC Joint Meeting, New Delhi, India · Nov 2023
The Microbiome and IBD: What do we know now?
Annual IBD Symposium, Egyptian Society of Crohn's and Colitis, Alexandria, Egypt · Sep 2023

All 22 talks and media items in the long-form CV →

Software, patents & service

AI systems & scientific software

I build usable AI and scientific systems, backed by 40+ public repositories. Selected work:

ChoreCodeIndependent AI evaluation / API
Independent AI evaluation research and engineering project: cost-aware evaluation of coding models on realistic maintenance work, with hidden tests, anti-gaming constraints, failure analysis, and a recommendations API.
MaLiAmPiNextflow / Python★ 22
Taxonomy-free harmonization & tokenization of raw microbiome data into stable, ML-ready features.
geneshotNextflow
Reference-free, gene-level metagenomics with rigorous dimensionality reduction.
Groups phylogenetic placements into stable phylogroups for downstream modeling.
sciluigiPython
Scientific-workflow wrapper around Spotify's Luigi.
arfPython
Algorithmic rRNA filtering for taxonomic identification.

Long-running public record across Python, Nextflow, Rust, JavaScript, data systems, and reproducible scientific workflows. Arctic Code Vault Contributor. · All repositories →

Patents

Phylogenetic placement using taxonomy-independent feature generation
US 2025/0037793 A1 · U.S. patent application (pending) · Filed Jul 2024 · published Jan 2025

Core method behind MaLiAmPi — generalizable, taxonomy-independent microbiome features via phylogenetic placement.

Leadership, testimony & service

Fleming Initiative — Data & AI/ML leadership
Imperial College London + Imperial College Healthcare NHS Trust · GSK founding partner

Leads and is accountable for GSK's AI/ML and data-engineering strategy and delivery across all six Grand Challenges. GSK scientific co-lead for Antibiotics and Antifungals; co-developed their scopes of work and scientific plans; established cross-organizational data coordination across industry and academic partners.

Expert declaration — COVID-19 in carceral settings
ACLU of Connecticut · federal litigation (2020)

Provided medical expert testimony on protecting incarcerated people from COVID-19.

Skills & education

AI and evaluation

Machine learning for biology

Applied biomedical AI and AI-for-Science strategyProtein and RNA language modelsLLMs and representation modelsDiffusion approaches for biological modelingUncertainty-aware biological-sequence modeling, propagated to downstream modelsDeep learning & predictive modelingViral evolution, infection timing & host–immune modelingEHR and multi-omics data

Evaluation & AI decision systems

Model evaluation, benchmarking & failure analysisBiomedical problem framing & method selectionUser discovery & workflow integrationInterface and API designUncertainty communication & appropriate useAdoption planning & operational evaluation

Responsible AI

Responsible AI and AI governanceClinical AI safety and model evaluationDeep learning for biodefenseDetection of anomalous AI use in health and biologySociotechnical risk in high-stakes systems

Software engineering

Python (data science & Django backend)PyTorch, TensorFlow, scikit-learn; pandas, NumPy, statsmodelsReproducible workflows: Nextflow; Slurm / PBS / SGESQL & NoSQL data modelingJavaScript / D3.js data visualizationC / C++ / R

Biology and clinical medicine

Clinical medicine

Transplant infectious diseaseImmunocompromised hosts & cancer-related infectionsInternal medicinePandemic critical care

Clinical trials, drug & vaccine development

Early-phase protocol development & trial biomarker analysisSite PI, global unblinded safety oversight & trial organizing committeesPreclinical R&D (vaccines & infectious disease)Development strategy & lifecycle managementAntimicrobial resistance (AMR)

Bench & computational biology

Microbiome, single-cell & multi-omics integrationBioinformatics pipeline developmentGene-therapy vector design (adeno-, lenti-, AAV)Organoid / in vitro modelingVaccinology & epitope selection

Leading people and programs

Leading teams

Building focused specialist ML teamsHiring, performance management & technical developmentCross-functional matrix leadership (20–30 people)Leading & delivering complex software systems

Programs & strategy

Cross-organizational scientific program designExecutive advising, delivery accountability & technical strategyTechnical and business cases for AI investmentBuild–buy–partner strategyVendor scoping & evaluationDurable operational handoff

Education

Transplant Infectious Diseases Fellowship — Joel Meyers Endowment Scholar
Fred Hutchinson Cancer Center · Seattle, WA · 2016 – 2017
Joel Meyers Endowment Scholarship, Fred Hutch Infectious Disease Sciences Program (2016, 2017): the named award for transplant infectious disease fellows, honoring the program’s founder and CMV pioneer Joel Meyers. One of 25 scholars named between 1996 and 2025.
Infectious Diseases Fellowship
University of Washington · Seattle, WA · 2013 – 2017
Internal Medicine Residency
University of Washington · Seattle, WA · 2011 – 2013
MD–PhD — Medical Scientist Training Program
University of Washington · Seattle, WA · 2001 – 2011
PhD, Pathology (stem-cell epigenetics)
Medical-school internal medicine at Madigan Army Medical Center; pediatrics clinical year at Seattle Children's Hospital.
BS — Biomedical Engineering & Computer Science (dual degree)
Johns Hopkins University · Baltimore, MD · 1997 – 2001

Outside work

Board games, photography, cycling, cross-country skiing, and writing.