The application of AI to biology is leading to revolutionary achievements with Alphafold for protein folding and generative models of artificial enzymes. However, a question remains open at a higher level of organization in cells, crucial for diagnosis and therapeutic strategies: how to predict the impact of multiple gene perturbations on biological functions? The aim of the project is to develop a deep learning approach leveraging on the one hand recent work on deep neural networks in the MILES team of Dauphine [1] and on the other hand experiments and mechanistic models developed at the ESPCI laboratory, showing that interactions between mutations are largely explained by non-linear but smooth functions of intermediate features [2-3]. How can we design a deep-learning model that can include these two steps of decomposition to learn gene interactions? For this purpose, the PhD will develop three complementary lines of research: (i) build deep nets on published datasets of combined genetic and drug perturbations on cells; (ii) computationally simulate the response of cells based on typical gene network architectures; (iii) apply the developed models to design, predict and interpret the outcome of large scale screening experiments done at ESPCI. The PhD will be co-supervised by Alexandre Allauzen, Professor at University Paris Dauphine and ESPCI (team Machine Intelligence and LEarning Systems) and Philippe Nghe (director of the Laboratory of Biophysics and Evolution, ESPCI Paris-PSL). It is funded by the CNRS. Possible profiles include former experience in deep learning, computational biology, physics, mathematics. The position is expected to start around September 2022. Contacts: alexandre.allauzen@dauphine.psl.eu; philippe.nghe@espci.psl.eu [1] H. Le, L. Vial, J. Frej, V. Segonne, M. Coavoux, B. Lecouteux, A. Allauzen, B. Crabbé, L. Besacier, Didier S. (2020) Flaubert: Unsupervised language model pre-training for French. Proceedings of the 12th Language Resources and Evaluation Conference [2] Kemble, H., Eisenhauer, C., Couce, A., Chapron, A., Magnan, M., Gautier, G., Nghe, P. & Tenaillon, O. (2020). Flux, toxicity, and expression costs generate complex genetic interactions in a metabolic pathway. Science advances, 6(23), eabb2236. [3] Nghe, P., Kogenaru, M., & Tans, S. J. (2018). Sign epistasis caused by hierarchy within signalling cascades. Nature communications, 9(1), 1-9. https://www.edpif.org/phorum/read.php?3,282,282#msg-282 Sat, 17 Oct 2026 17:01:17 +0200 Phorum 5.2.23 https://www.edpif.org/phorum/read.php?3,282,282#msg-282 https://www.edpif.org/phorum/read.php?3,282,282#msg-282
For this purpose, the PhD will develop three complementary lines of research: (i) build deep nets on published datasets of combined genetic and drug perturbations on cells; (ii) computationally simulate the response of cells based on typical gene network architectures; (iii) apply the developed models to design, predict and interpret the outcome of large scale screening experiments done at ESPCI.

The PhD will be co-supervised by Alexandre Allauzen, Professor at University Paris Dauphine and ESPCI (team Machine Intelligence and LEarning Systems) and Philippe Nghe (director of the Laboratory of Biophysics and Evolution, ESPCI Paris-PSL). It is funded by the CNRS. Possible profiles include former experience in deep learning, computational biology, physics, mathematics. The position is expected to start around September 2022.
Contacts: alexandre.allauzen@dauphine.psl.eu; philippe.nghe@espci.psl.eu

[1] H. Le, L. Vial, J. Frej, V. Segonne, M. Coavoux, B. Lecouteux, A. Allauzen, B. Crabbé, L. Besacier, Didier S. (2020) Flaubert: Unsupervised language model pre-training for French. Proceedings of the 12th Language Resources and Evaluation Conference
[2] Kemble, H., Eisenhauer, C., Couce, A., Chapron, A., Magnan, M., Gautier, G., Nghe, P. & Tenaillon, O. (2020). Flux, toxicity, and expression costs generate complex genetic interactions in a metabolic pathway. Science advances, 6(23), eabb2236.
[3] Nghe, P., Kogenaru, M., & Tans, S. J. (2018). Sign epistasis caused by hierarchy within signalling cascades. Nature communications, 9(1), 1-9.]]> pnghe Thèses hors de l'EDPIF Thu, 28 Apr 2022 20:10:19 +0200