Uncertainty Quantification and Targeted Molecular Simulation for Generative Protein Engineering

Internship subject

Subject 2 — PFE-02

Measure where a protein model is truly uncertain, and spend expensive physics simulation only in those areas.

What you’ll build

• Uncertainty estimator: Bayesian Monte Carlo ensembles over ESM-2.
• Targeted simulation: OpenMM / PyRosetta on risky regions, with ΔG readout.
• Screening module: Check against regulated-sequence registries.
• Retrospective benchmark: Use ProteinGym to assess whether the approach would have caught known failures.

Tech stack

Python, OpenMM, PyTorch, PyRosetta, Transformers, Biopython, ESM-2 and SciPy.

Key challenge

Separate what the model does not know from genuine biological variation.

Target

Achieve at least 40% less simulation compute at equal or better ranking quality.

Profile and context

• Profile: Biotech engineer or master.
• Format: 1 intern, for 4 to 6 months.
• Market: Global.
• Full-time, on-site internship in Monastir, Tunisia.
• Candidates should be willing to learn and take on the challenge, have relevant skills and interests, possess an adequate level of written and spoken English, and be able to work alongside people from different backgrounds and disciplines.
• Advantages include a monthly bonus according to profile, an achievement bonus and in-demand training.