Artificial Intelligence Researcher

AI scientist for computational protein design - #AI4good

A post-doctoral position in AI (Deep Learning) and computational protein design [1] is available at MIA Toulouse (miat.inrae.fr), France. MIAT is a lab of the MathNum department of the French institute for agriculture, food, and the environment.

Context: INRAE has launched a deep-tech research initiative looking for disruptive results and a high societal and scientific impact. A multidisciplinary team of experts regrouping AI-computer scientists, molecular modeling/biotechnology scientists, and virologists has been gathered to answer this call, based on the joint experience of several of its members in developing new AI-based computational protein design tools and applying them to real-world targets. Our tools have already shown their capacities on several proof-of-concept experiments, leading to improved enzymes, novel nanobodies or small protein scaffolds for diagnosis and viral neutralization, and self-assembling proteins. The INRAE-funded project aims to build new highly efficient and precise generative AI architectures for designing new proteins with high impact against selected viral targets.

Position: The researcher will play a key role in this multidisciplinary project. He/she will be responsible for developing, training, and testing AI architectures and designing pipelines that can reliably generate and evaluate protein structures and sequences with specific functions. This research will be conducted in tight collaboration with AI scientists, computational biologists, biochemists, and virologists for experimental validation.

This recruitment will be carried out as a two-year fixed-term contract, renewable for one year, funded by INRAE, in Toulouse (France). It can start as soon as Q2-3 2025.

Expected skills: We are looking for a dynamic scientist, highly motivated by the development of methods that can contribute to meeting some of the major challenges of the world of tomorrow. The ideal candidate will be proficient in Python and Torch (ideally), familiar with machine & deep learning principles and tools, with a good knowledge of deep learning architectures, including GNNs, generative architectures such as DDPMs and variants, Transformers, xLSTM, ... from their principles and implementations to effective application to real-world problems that may include non-trivial equivariances. Experience in developing deep learning architectures, constructing training sets, and training DL models on multi-GPU infrastructures is a big plus. Further expertise in bioinformatics, structural biology, or biophysics would be a plus, but is not a requirement.

Application: applicants should send, as soon as possible, a detailed curriculum vitae with Git repos URLs, a letter of intent explaining their motivations for the position, and the contact details of at least two references to Thomas Schiex <[email protected]> and Sophie Barbe <[email protected]>.

[1] Albanese KI, Barbe S, Tagami S, Woolfson DN, Schiex T. Computational protein design. Nature Reviews Methods Primers. 2025 Feb 27;5(1):13. Accessible at https://rdcu.be/ebvNU