Matthieu Chaldebas

ML Scientist · Genomics

Building end-to-end ML frameworks to decode how genetic variants drive disease

13 Publications
263+ Citations
8 h-index
5 International Talks
409K WGS Participants

Published in Nature · Science Immunology · PNAS · Immunity · AJHG · J. Clin. Investigation · J. Exp. Medicine

Presenting at ISMB 2026 — Washington, DC

Seeking postdoctoral & industry research positions · Available 2027

Email Me LinkedIn GitHub Scholar ORCID Download CV

Professional Experience

I build ML frameworks that decode what non-coding genetic variants mechanistically do — not just whether evolution preserved them. My work spans from model training on population-scale WGS data to rare-variant discovery, target identification, and multi-omic validation in patient cohorts.

Computational Biologist

St. Giles Laboratory of Human Genetics of Infectious Diseases

The Rockefeller University, New York & Imagine Institute, Paris | 2021 – Present

PhD Candidate

2022 – Present

  • Engineered and deployed 5ULTRA, a Python ML framework combining SpliceAI, uORF databases, and a Random Forest classifier — achieving AUC 0.82 vs. 0.75 (CADD) on independent ClinVar benchmarks; validated predictions with Olink proteomics (n = 46,362, cis-pQTL effect sizes >5× baseline); published 1st author in AJHG (2026).
  • Applied 5ULTRA to identify novel disease-causing mechanisms in cancer (ABI1, NRAS) and infectious disease (TNF, RPSA), enabling mechanistic target identification at genome-wide scale.
  • Conducted a population-scale rare-variant burden study in 408,423 UK Biobank WGS participants across 59 quantitative traits — achieving a 1.6× improvement over CADD and identifying 58 genome-wide significant associations including 18 novel targets (manuscript in preparation).

Bioinformatics Assistant

2021 – 2022

  • Built and maintained germline variant-calling pipelines across 26,000+ patient WES/WGS datasets (GATK, VEP, Snakemake on HPC), enabling disease-gene discoveries published in Nature, Science Immunology, and Immunity.
  • Performed bulk and single-cell RNA-seq analyses (differential expression, pathway enrichment, Seurat) across multiple disease cohorts.
  • Co-developed AGAIN, a genome-wide intronic splicing variant detector, contributing key features and WES/WGS validation (PNAS, 2023).

Data Scientist Intern

BioMérieux, Lyon, France | 6 Months – 2021

  • Developed supervised Lasso ML models for antibiotic resistance prediction from microbial genomic data, extending cross-validation to account for complex phylogenetic structures.
  • Implemented novelty detection algorithms to clean and standardize heterogeneous clinical datasets, improving downstream model reliability.

Earlier Internships

CEA Paris Saclay (4 months, 2019) · Imagine Institute, Paris (2 months, 2018)

  • Built bioinformatics data pipelines for phylogenetic analysis and 3D protein modeling (CEA).
  • Western blot and RT-PCR assays to evaluate drug effects on gene expression (Imagine Institute).

Publications

Selected for Platform Presentation — ASHG 2025
Published in AJHG (April 2026)

Genome-wide detection of human 5’UTR variants that impact protein translation

Chaldebas M, Ponsin K, Bohlen J, Conil C, Mourelatos H, Stenson PD, Cooper DN, Abel L, Casanova JL, Cobat A#, Zhang P#. (#Equal contribution)
The American Journal of Human Genetics (AJHG), 2026.

Research Highlight: Presented as a Platform Talk at ASHG 2025. We developed 5ULTRA, a Random Forest classifier integrating SpliceAI and uORF biology to prioritize 5'UTR variants — achieving AUC 0.82 vs. 0.75 (CADD) on independent ClinVar benchmarks. Validated predictions using Olink proteomics (n = 46,362): cis-pQTL variants predicted by 5ULTRA carry effect sizes >5× larger than baseline. Applied to identify disease-causing mechanisms in cancer (ABI1, NRAS) and infectious diseases (TNF, RPSA).

Read at AJHG Try the Webserver Code on GitHub
Selected for Platform Presentation — ASHG 2026
First Author · In Preparation · 2026

Mechanistic 5'UTR Variant Scoring Expands Rare-Variant Discovery and Reveals Translational Dosage Control in the UK Biobank

Chaldebas M, Ponsin K, Mourelatos HA, Seeleuthner Y, Conil C, Casanova JL, Zhang P#, Cobat A#. (#Co-senior & corresponding authors)

Research Highlight: Applied 5ULTRA to 408,423 UK Biobank WGS participants (non-Finnish European discovery cohort) across 59 quantitative traits — achieving a 1.6× improvement over CADD (58 vs. 37 genome-wide significant associations, 34 exclusive to mechanistic annotation). Ablating 5ULTRA-annotated variants from CADD eliminated 27% of its signal, locating that power in uORF and Kozak architecture. Among 18 novel discoveries absent from a concurrent UKB 5'UTR catalog, NELFCD reached −log₁₀P = 49.6 for platelet distribution width. The 5'UTR behaves as a directional, predominantly repressor-driven translational dosage layer, and 5'UTR loss-of-function burden effects are quantitatively comparable to exome protein-truncating variants (r = 0.68, P = 3.7 × 10⁻⁴). Rare 5ULTRA variants were enriched 3.3-fold among the top decile of cis-pQTL signals (493 genes; Hawkes et al., 2025), and associations replicated directionally across non-European ancestries (89% sign-concordant, r = 0.84).

Code repository in preparation
Co-Second Author · PNAS 2023

Genome-wide detection of human intronic AG-gain variants located between splicing branchpoints and canonical splice acceptor sites

Zhang P, Chaldebas M*, Ogishi M*, et al. (*Equal contributions)
PNAS, 2023.

My Contribution: As co-second author (equal contributions), I developed key algorithmic features and performed WES/WGS validation of variant impacts at genome-wide scale.

Read at PNAS Try the Webserver Code on GitHub

Collaborative Contributions

My bioinformatics work at The Rockefeller University directly supported gene discovery across the lab's portfolio in infectious disease and immunology — WES/WGS variant calling, bulk and single-cell RNA-seq, and custom pipeline development across cohort studies spanning thousands of patients.

Publications supported by this work:

263+ Total Citations on Google Scholar

Conferences & Talks

2026

ISMB 2026

Washington, DC, USA

Talk

5ULTRA: genome-wide prioritization of 5'UTR variants impacting protein translation.

2026

ASHG 2026 Annual Meeting

Montréal, QC, Canada

Platform Presentation

Talk — Mechanistic 5'UTR variant scoring expands rare-variant discovery and reveals translational dosage control in the UK Biobank.

2026

Webinaire Régulome — bioinfo-diag.fr

Online Webinar

Invited talk — 5ULTRA: genome-wide prioritization of 5'UTR variants impacting protein translation.

2025

ASHG 2025 Annual Meeting

Boston, MA, USA

Platform Presentation

Talk — 5ULTRA: genome-wide prioritization of 5'UTR variants impacting protein translation.

2025

Splicing 2025

Caparica, Lisbon, Portugal

Presentation — AGAIN: genome-wide detection of intronic AG-gain splicing variants.

Technical Skills

Machine Learning & Programming

Machine Learning Python R Bash AlphaGenome PHP, HTML & CSS

Genomics & Multi-Omics

WES / WGS Analysis Rare-Variant Burden (REGENIE) RNA-Seq (Bulk & scRNA) Proteomics (Olink / pQTL) SpliceAI GWAS / Statistical Genetics 5'UTR Biology Population Databases

Engineering & Infrastructure

Version Control (Git) Cluster Computing (HPC/SLURM) DNAnexus / UKB RAP Snakemake Docker / Singularity Conda

Education

PhD in Bioinformatics

Université Paris Cité, France | 2022 – December 2026 (Expected)

Thesis: ML frameworks for genome-wide 5'UTR and splicing variant interpretation in human disease.

Joint program: The Rockefeller University (New York) & Imagine Institute (Paris). Advisors: Drs. Casanova, Cobat & Zhang.

M.Sc. in Biotechnology Engineering

Sup'Biotech, Paris, France | 2016 – 2021

Specialization in health data science.

Exchange semester — Immunology & Human Disease, UC San Diego, La Jolla, CA.

Languages

French — Native  ·  English — Fluent