Gabriel Cretin

I build deep learning models for protein biology and the infrastructure to train them. From designing neural architectures to managing GPU clusters, I handle both the science and the systems.

Get in touch

Download CV Updated: Jan 2026

research ⟷ infrastructure

Deep Learning

Computational Biologist - Ph.D

Specialized in Protein Language Models and distributed GPU training using PyTorch with 40k+ A100/H100 GPU hours on national HPC (IDRIS/CNRS). Published research on protein structure analysis, prediction, and generative modeling.

Input pLM Embeddings

Training Model

3D Structure

Flexibility

Fold Recognition

PyTorch Lightning ESM-2 / Ankh Autoencoders Contrastive Learning

Explore research

SRE & MLOps

Linux System Administrator

Running 15+ production web services including GitLab, JupyterHub, and structural biology tools. Managing a GPU cluster with 20+ workstations, 380+ CPUs, and 1 PB+ backup infrastructure.

gabriel@hpc-master:~

➜ ~ █

Ansible Docker Swarm SLURM / HPC OpenLDAP

View infrastructure

Professional Profile

A rare hybrid profile combining 6 years of Linux infrastructure management with 4 years of deep learning research. I design novel AI architectures and deploy them on the infrastructure I build, training on national supercomputers to production APIs serving 18K+ weekly requests.

40k+

GPU hours (IDRIS/CNRS)

8

Peer-reviewed papers

700+

HPC cores managed

18K+

Weekly API requests

Scientific Track

Deep Learning Research

Focus on Protein Language Models (ESM-2, Ankh, ProtTrans) and generative architectures. Developed Adversarial Autoencoders for embedding compression and contrastive learning for improved fold recognition, published in top-tier journals.

8 peer-reviewed publications
40k+ GPU hours on IDRIS/CNRS
4 production web tools (PYTHIA, SWORD2, ICARUS, PEGASUS)

Engineering Track

SRE & Infrastructure

Managing a complete Linux ecosystem: web servers, GPU clusters, HPC, and centralized authentication. Building reliable platforms for research teams with automated provisioning and monitoring.

14 web servers, 12 databases (~18K views/week)
HPC cluster: 48 nodes, 708 cores
1 PB+ cumulated storage infrastructure

AI & Protein Science

Ph.D Research (2021–2025)

Representation Learning & Generative Modeling

Designed Adversarial Autoencoder (AAE) architectures to compress high-dimensional pLM embeddings (ESM-2, Ankh) into fixed-size latent spaces. Implemented contrastive triplet learning to improve structural fold recognition, surpassing state-of-the-art structure-based methods. Explored de novo protein design through latent space interpolation.

Papers in preparation - currently writing manuscripts on representation learning, adversarial autoencoders, and contrastive learning.

Adversarial Autoencoders Triplet Loss ESM-2 / Ankh / ProtTrans Fold Recognition

Tech Stack

ML Engineering

PyTorch / Lightning Expert
Python / Bash Expert
HPC / SLURM Advanced
AlphaFold / Foldseek Advanced

SRE & Infrastructure

HPC & Compute

Cluster Management

5 GPU servers for deep learning & molecular dynamics. SGI HPC cluster: 48 nodes, 708 cores. 25 Linux workstations with 700+ CPUs.

SLURM Spack / Environment Modules

Services

Self-Hosted Stack

14 web servers, 2 APIs, 12 databases serving ~18K views/week. GitLab, Mattermost, JupyterHub, centralized auth.

GitLab CI/CD OpenLDAP JupyterHub FastAPI

Automation & Networking

Infrastructure as Code

1 PB+ backup infrastructure (3 dedicated servers). Ansible playbooks, Docker Swarm orchestration, Samba/NFS storage.

Ansible Docker Swarm Apache2 / NFS EfficientIP

Publications

Featured Publications

PEGASUS

Protein Science 2025

Co-first author

Protein flexibility is essential to its biological function. However, experimental methods for its assessment, such as X‐ray crystallography and nucle...

DOI

Protein flexibility is essential to its biological function. However, experimental methods for its assessment, such as X‐ray crystallography and nuclear magnetic resonance spectroscopy, are often limited by experimental variability and high cost, leading to a gap between the number of identified protein sequences and the available experimental information on protein dynamics. On the other hand, molecular dynamics (MD) simulations provide a uniform and detailed description of the expected protein flexibility, and the availability and quality of such data are increasing significantly during the last years. In this study, we use the recently released ATLAS database to develop ProtEin lanGuAge models for prediction of SimUlated dynamicS (PEGASUS), a sequence‐based predictor of MD‐derived information on protein flexibility (https://dsimb.inserm.fr/PEGASUS). PEGASUS integrates four different representations of protein sequences generated by Protein Language Models to predict residue‐wise MD‐derived values of backbone fluctuation (root mean square fluctuation), Phi and Psi dihedral angles standard deviation, and average Local Distance Difference Test across the trajectory. The PEGASUS web server was optimized to perform instantaneous predictions for an individual protein sequence and also allows batch submission of up to 100 sequences of 1 k residues each. For more complex queries, we also release PEGASUS as a user‐friendly standalone utility (https://github.com/DSIMB/PEGASUS).

ATLAS

Nucleic Acids Research 2024

Contributing author

Dynamical behaviour is one of the most crucial protein characteristics. Despite the advances in the field of protein structure resolution and predicti...

DOI

Dynamical behaviour is one of the most crucial protein characteristics. Despite the advances in the field of protein structure resolution and prediction, analysis and prediction of protein dynamic properties remains a major challenge, mostly due to the low accessibility of data and its diversity and heterogeneity. To address this issue, we present ATLAS, a database of standardised all-atom molecular dynamics simulations, accompanied by their analysis in the form of interactive diagrams and trajectory visualisation. ATLAS offers a large-scale view and valuable insights on protein dynamics for a large and representative set of proteins, by combining data obtained through molecular dynamics simulations with information extracted from experimental structures. Users can easily analyse dynamic properties of functional protein regions, such as domain limits (hinge positions) and residues involved in interaction with other biological molecules. Additionally, the database enables exploration of proteins with uncommon dynamic properties conditioned by their environment such as chameleon subsequences and Dual Personality Fragments. The ATLAS database is freely available at https://www.dsimb.inserm.fr/ATLAS.

SWORD2

Nucleic Acids Research 2022

First author

Understanding the functions and origins of proteins requires splitting these macromolecules into fragments that could be independent in terms of folding, activity, or evolution. For that purpose, stru

DOI

Understanding the functions and origins of proteins requires splitting these macromolecules into fragments that could be independent in terms of folding, activity, or evolution. For that purpose, structural domains are the typical level of analysis, but shorter segments, such as subdomains and supersecondary structures, are insightful as well. Here, we propose SWORD2, a web server for exploring how an input protein structure may be decomposed into ‘Protein Units’ that can be hierarchically assembled to delimit structural domains. For each partitioning solution, the relevance of the identified substructures is estimated through different measures. This multilevel analysis is achieved by integrating our previous work on domain delineation, ‘protein peeling’ and model quality assessment. We hope that SWORD2 will be useful to biologists searching for key regions in their proteins of interest and to bioinformaticians building datasets of protein structures. The web server is freely available online: https://www.dsimb.inserm.fr/SWORD2.

MEDUSA

Journal of Molecular Biology 2021

Contributing author

MEDUSA: Prediction of Protein Flexibility from Sequence

DOI

Papers in Preparation

Currently writing manuscripts on my Ph.D. research — expected submission in 2025:

Representation Learning with Adversarial Autoencoders — Compressing pLM embeddings into continuous latent spaces for protein generation
Contrastive Learning for Fold Recognition — Triplet-based training to improve structural similarity detection beyond structure-based methods

Complete Bibliography

10 peer-reviewed papers • 4 first/co-first author • Updated: 2026-01-28

2025

Ragousandirane Radjasandirane, Gabriel Cretin, Julien Diharce, Alexandre G. de Brevern, Jean-Christophe Gelly. PATHOS: Predicting Variant Pathogenicity by Combining Protein Language Models and Biological Features.

DOI:10.64898/2025.12.22.25342839|

Predicting the pathogenic impact of missense variants is essential for understanding and diagnosing genetic diseases. These approaches have undergone significant evolution, with the latest methodologies based on deep learning approaches. Nonetheless, only a limited number use the potential of Protein Language Models (PLMs), which have demonstrated strong performance across various protein-related tasks. A new predictor, called PATHOS, was developed; it combines embeddings from an optimal set of two PLMs, namely ESM C 600M and Ankh 2 Large. Their embeddings were combined with additional crucial biological features such as phylogenetic probabilities, allele frequency, and protein annotations; they were aggregated using a fully connected layer architecture. Compared to 65 other predictors on clinical data, PATHOS outperforms state-of-the-art performance. It achieves a Matthews Correlation Coefficient (MCC) of 0.591 on a manually and carefully curated clinical dataset and 0.826 on a ClinVar dataset, surpassing other leading tools. Furthermore, case studies on the progesterone receptor and the KCNQ1 ion channel illustrate that PATHOS can identify functionally critical regions and known pathogenic mutations missed by other leading predictors like AlphaMissense. To ensure broad accessibility and facilitate use by non-specialists, a user-friendly web server containing a database of 140 millions precomputed predictions from human protein from Swiss-Prot was provided. The web server is available at: https://dsimb.inserm.fr/PATHOS/

Yann Vander Meersche, Gabriel Duval, Gabriel Cretin, Aria Gheeraert, Jean‐Christophe Gelly, Tatiana Galochkina. PEGASUS: Prediction of MD‐derived protein flexibility from sequence. Protein Science, 2025.

DOI:10.1002/pro.70221|Co-first author

Protein flexibility is essential to its biological function. However, experimental methods for its assessment, such as X‐ray crystallography and nuclear magnetic resonance spectroscopy, are often limited by experimental variability and high cost, leading to a gap between the number of identified protein sequences and the available experimental information on protein dynamics. On the other hand, molecular dynamics (MD) simulations provide a uniform and detailed description of the expected protein flexibility, and the availability and quality of such data are increasing significantly during the last years. In this study, we use the recently released ATLAS database to develop ProtEin lanGuAge models for prediction of SimUlated dynamicS (PEGASUS), a sequence‐based predictor of MD‐derived information on protein flexibility (https://dsimb.inserm.fr/PEGASUS). PEGASUS integrates four different representations of protein sequences generated by Protein Language Models to predict residue‐wise MD‐derived values of backbone fluctuation (root mean square fluctuation), Phi and Psi dihedral angles standard deviation, and average Local Distance Difference Test across the trajectory. The PEGASUS web server was optimized to perform instantaneous predictions for an individual protein sequence and also allows batch submission of up to 100 sequences of 1 k residues each. For more complex queries, we also release PEGASUS as a user‐friendly standalone utility (https://github.com/DSIMB/PEGASUS).

Aria Gheeraert, Thomas Bailly, Yani Ren, Ali Hamraoui, Julie Te, Yann Vander Meersche, Gabriel Cretin, Ravy Leon Foun Lin, Jean-Christophe Gelly, Serge Pérez, Frédéric Guyon, Tatiana Galochkina. DIONYSUS: a database of protein–carbohydrate interfaces. Nucleic Acids Research, 2025.

DOI:10.1093/nar/gkae890|

Protein-carbohydrate interactions govern a wide variety of biological processes and play an essential role in the development of different diseases. Here, we present DIONYSUS, the first database of protein-carbohydrate interfaces annotated according to structural, chemical and functional properties of both proteins and carbohydrates. We provide exhaustive information on the nature of interactions, binding site composition, biological function and specific additional information retrieved from existing databases. The user can easily search the database using protein sequence and structure information or by carbohydrate binding site properties. Moreover, for a given interaction site, the user can perform its comparison with a representative subset of non-covalent protein-carbohydrate interactions to retrieve information on its potential function or specificity. Therefore, DIONYSUS is a source of valuable information both for a deeper understanding of general protein-carbohydrate interaction patterns, for annotation of the previously unannotated proteins and for such applications as carbohydrate-based drug design. DIONYSUS is freely available at www.dsimb.inserm.fr/DIONYSUS/.

Charlotte Perin, Gabriel Cretin, Jean-Christophe Gelly. Hierarchical Analysis of Protein Structures: From Secondary Structures to Protein Units and Domains. Methods in Molecular Biology, 2025.

DOI:10.1007/978-1-0716-4213-9_18

2024

Yann Vander Meersche, Gabriel Cretin, Aria Gheeraert, Jean-Christophe Gelly, Tatiana Galochkina. ATLAS: protein flexibility description from atomistic molecular dynamics simulations. Nucleic Acids Research, 2024.

DOI:10.1093/nar/gkad1084|

Dynamical behaviour is one of the most crucial protein characteristics. Despite the advances in the field of protein structure resolution and prediction, analysis and prediction of protein dynamic properties remains a major challenge, mostly due to the low accessibility of data and its diversity and heterogeneity. To address this issue, we present ATLAS, a database of standardised all-atom molecular dynamics simulations, accompanied by their analysis in the form of interactive diagrams and trajectory visualisation. ATLAS offers a large-scale view and valuable insights on protein dynamics for a large and representative set of proteins, by combining data obtained through molecular dynamics simulations with information extracted from experimental structures. Users can easily analyse dynamic properties of functional protein regions, such as domain limits (hinge positions) and residues involved in interaction with other biological molecules. Additionally, the database enables exploration of proteins with uncommon dynamic properties conditioned by their environment such as chameleon subsequences and Dual Personality Fragments. The ATLAS database is freely available at https://www.dsimb.inserm.fr/ATLAS.

2023

Gabriel Cretin, Charlotte Périn, Nicolas Zimmermann, Tatiana Galochkina, Jean-Christophe Gelly, Lenore Cowen. ICARUS: flexible protein structural alignment based on Protein Units. Bioinformatics, 2023.

DOI:10.1093/bioinformatics/btad459|First author

Motivation Alignment of protein structures is a major problem in structural biology. The first approach commonly used is to consider proteins as rigid bodies. However, alignment of protein structures can be very complex due to conformational variability, or complex evolutionary relationships between proteins such as insertions, circular permutations or repetitions. In such cases, introducing flexibility becomes useful for two reasons: (i) it can help compare two protein chains which adopted two different conformational states, such as due to proteins/ligands interaction or post-translational modifications, and (ii) it aids in the identification of conserved regions in proteins that may have distant evolutionary relationships. Results We propose ICARUS, a new approach for flexible structural alignment based on identification of Protein Units, evolutionarily preserved structural descriptors of intermediate size, between secondary structures and domains. ICARUS significantly outperforms reference methods on a dataset of very difficult structural alignments. Availability and implementation Code is freely available online at https://github.com/DSIMB/ICARUS.

2022

Gabriel Cretin, Tatiana Galochkina, Yann Vander Meersche, Alexandre G de Brevern, Guillaume Postic, Jean-Christophe Gelly. SWORD2: hierarchical analysis of protein 3D structures. Nucleic Acids Research, 2022.

DOI:10.1093/nar/gkac370|PMC9252838|First author

Understanding the functions and origins of proteins requires splitting these macromolecules into fragments that could be independent in terms of folding, activity, or evolution. For that purpose, structural domains are the typical level of analysis, but shorter segments, such as subdomains and supersecondary structures, are insightful as well. Here, we propose SWORD2, a web server for exploring how an input protein structure may be decomposed into ‘Protein Units’ that can be hierarchically assembled to delimit structural domains. For each partitioning solution, the relevance of the identified substructures is estimated through different measures. This multilevel analysis is achieved by integrating our previous work on domain delineation, ‘protein peeling’ and model quality assessment. We hope that SWORD2 will be useful to biologists searching for key regions in their proteins of interest and to bioinformaticians building datasets of protein structures. The web server is freely available online: https://www.dsimb.inserm.fr/SWORD2.

Nora El Jahrani, Gabriel Cretin, Alexandre G. de Brevern. CALR-ETdb, the database of calreticulin variants diversity in essential thrombocythemia. Platelets, 2022.

DOI:10.1080/09537104.2020.1869712

2021

Gabriel Cretin, Tatiana Galochkina, Alexandre de Brevern, Jean-Christophe Gelly. PYTHIA: Deep Learning Approach for Local Protein Conformation Prediction. International Journal of Molecular Sciences, 2021.

DOI:10.3390/ijms22168831|PMID:34445537|PMC8396346|First author

Protein Blocks (PBs) are a widely used structural alphabet describing local protein backbone conformation in terms of 16 possible conformational states, adopted by five consecutive amino acids. The representation of complex protein 3D structures as 1D PB sequences was previously successfully applied to protein structure alignment and protein structure prediction. In the current study, we present a new model, PYTHIA (predicting any conformation at high accuracy), for the prediction of the protein local conformations in terms of PBs directly from the amino acid sequence. PYTHIA is based on a deep residual inception-inside-inception neural network with convolutional block attention modules, predicting 1 of 16 PB classes from evolutionary information combined to physicochemical properties of individual amino acids. PYTHIA clearly outperforms the LOCUSTRA reference method for all PB classes and demonstrates great performance for PB prediction on particularly challenging proteins from the CASP14 free modelling category.

Yann Vander Meersche, Gabriel Cretin, Alexandre G. de Brevern, Jean-Christophe Gelly, Tatiana Galochkina. MEDUSA: Prediction of Protein Flexibility from Sequence. Journal of Molecular Biology, 2021.

DOI:10.1016/j.jmb.2021.166882

CV / Resume

Download the PDF and browse a short timeline snapshot.

Download CV Updated: Jan 2026

Oct 22, 2025
PhD Thesis Defense

Université Paris Cité - "Deep learning approaches for protein analysis, prediction, and generation."
Sep 2021 – Oct 2025
PhD Student

DSIMB Lab - Protein Language Models embeddings compression into continuous latent space for generation,
and protein structure analysis and prediction
Sep 2019 – Present
Lead Linux System Administrator

Managing the full-stack infrastructure: 30+ GPU workstations, GPU cluster, 1 PB+ storage and containerized web services.
2017 – 2019
MSc (Master) - Biology, Computer Science, Bioinformatics

Université Paris Cité - with honors (rank 2/23).
2016 – 2017
BSc (Licence 3)- Biology, Computer Science, Bioinformatics

Université Paris Cité - rank 6/21.
2014 – 2016
Two-year degree (D.U.T) - Bioengineering & Bioinformatics

Université de Clermont-Ferrand (Campus Aurillac) - ranks 4/45 (1st year) and 5/34 (2nd year).

Contact

Best way to reach me: gabrielcretin@gmail.com

Paris, France

🇫🇷 French • 🇬🇧 English

GitHub LinkedIn DSIMB

Gabriel Cretin

Computational Biologist - Ph.D

Linux System Administrator

Professional Profile

Scientific Track

Deep Learning Research

Engineering Track

SRE & Infrastructure

AI & Protein Science

Representation Learning & Generative Modeling

ML Engineering

SRE & Infrastructure

Cluster Management

Self-Hosted Stack

Infrastructure as Code

Publications

Featured Publications

PEGASUS

ATLAS

SWORD2

MEDUSA

Papers in Preparation

Complete Bibliography

2025

2024

2023

2022

2021

CV / Resume

PhD Thesis Defense

PhD Student

Lead Linux System Administrator

MSc (Master) - Biology, Computer Science, Bioinformatics

BSc (Licence 3)- Biology, Computer Science, Bioinformatics

Two-year degree (D.U.T) - Bioengineering & Bioinformatics

Contact