Stage M2 - Deep Learning for mapping the structural space of microproteins

SFBI; Gif-sur-Yvette; On-site; EUR 12,000 - 18,000; Part time

Benefits offered by this job

HPC/GPU resources

Mentorship from researchers

Job summary

SFBI in Île-de-France seeks an M2 internship to map the structural space of microproteins using deep learning. You will work with ESMFlow and MD trajectories to predict stable conformations from noncoding sequences, advancing understanding of these small proteins.

The project blends ML, molecular dynamics, and evolution, with access to HPC/GPU resources and close supervision from researchers. Ideal for early-career scientists.

Qualifications

• Programming in Python is mandatory.
• Interest in machine learning and structural biology is essential.
• MD simulations or protein-structure analysis experience is a plus.

Responsibilities

• Learn from MD to identify stable conformations and assess persistence.
• Extract descriptors of conformational persistence and train ML models.
• Rank candidate structures for microproteins arising from noncoding sequences.
• Explore protein-language models or structural embeddings if progress allows.

Skills

Python

Machine learning

Structural biology

Education

M2 / Master 2 in Computational Biology

Tools

MD simulations

Protein structure analysis

Job description

Stage M2 - Deep Learning for mapping the structural space of microproteins

deep learning microproteins structure prediction noncoding genome

Description

Context: Genomes are traditionally divided into coding and noncoding regions, yet noncoding regions are pervasively translated and give rise to thousands of small proteins, called microproteins, encoded by short open reading frames. Some become established as new genes, illustrating how noncoding regions act as a source of molecular innovation; others are involved in immunity, stress response or cancer. Yet a fundamental question remains largely unanswered: what do these proteins actually look like?

Preliminary results: Precisely, these microproteins emerging from noncoding regions have undergone little or no selection at the protein level. Current deep-learning structure predictors often struggle with these short, poorly conserved sequences. We recently identified an intriguing class of microproteins with globular-like amino-acid composition but unusual sequence organization. Using ESMFlow, we found that they are associated with folding potential with stable secondary structures but sample a broad diversity of global conformations. Molecular-dynamics (MD) simulations support this picture and distinguish conformations that remain persistent from those that rapidly unfold. We have started testing these predictions experimentally and revealed different behaviors, ranging from small species to heterogeneous or well-defined oligomeric assemblies. Strikingly, negative-stain electron microscopy and ongoing cryo-EM characterization further demonstrate that these unusual sequences can form organized 3D structures.

Hypothesis: Together, these results suggest that microproteins populate rich and heterogeneous structural landscapes. However, mapping these landscapes experimentally or by MD one protein at a time is far too costly to scale to the thousands of microproteins now being discovered.

Objectives: The aim of this internship is to learn from molecular dynamics which predicted conformations are stable and which are not.

The student will work with conformational ensembles generated by ESMFlow and MD trajectories already available in the team. They will extract descriptors of conformational persistence and train and evaluate machine-learning models able to rank candidate structures for these intriguing microproteins arising from noncoding and therefore unevolved sequences. Depending on progress, protein-language-model or structural embeddings may also be explored. The project may also be extended towards generating new MD data or applying the model at a larger scale to the thousands of microproteins detected in our Ribosome Profiling data.

The longer-term goal is to develop a fast computational tool able to identify plausible conformations for thousands of poorly conserved or recently emerged proteins involve in stress response, adaptation or cancers.

Positioning: This internship lies at the interface of machine learning, MD simulations and molecular evolution, and addresses a genuinely open problem: there is currently no general method designed to describe the structural behavior of sequences that evolution has not yet shaped.

Environnement: The internship will take place in the Computational Biology of Noncoding Genomes team at I2BC and will be co-supervised by our collaborator Samuel Murail (Univ Paris-Cité). Our team brings together complementary expertise in structural bioinformatics, genomics, OMICS analyses and molecular modelling, and includes many young researchers (PhD students, engineers), providing a dynamic environment for early-career scientists. We are located in the new buildings of the I2BC, south of Paris, on the historic and green campus of the CNRS. The I2BC is a recent and highly dynamic institute with around 700 people, including about 150 PhD students and postdoctoral researchers. The student will benefit from our HPC/GPU resources and close interactions with structural biologists who will experimentally characterize selected microproteins.

Required Technical Skills: Programming in Python (mandatory). An interest in machine learning and in structural biology is essential. Previous experience with molecular dynamics or with protein structure analysis is a plus but is not required

Contacts

Samuel Murail
[e-mail]