Gianni De Fabritiis

Gianni De Fabritiis

Universitat Pompeu Fabra

Experimental Sciences & Mathematics

ICREA research professor, associate professor, and group leader of the Computational Science Laboratory at UPF, and CEO/CSO at Acellera Ltd. Bachelor's degree with honors in applied mathematics (1997) from the University of Bologna and a Ph.D. from the University of London (2002). I worked for the CINECA supercomputing center in Italy (1998-1999) and was a  postdoctoral researcher at University College London (2003-2006). In 2006, I founded Acellera Ltd where I currently act as a CSO. In 2008 I won a tenure-track Ramon y Cajal research position and later the national I3-tenured program. In 2014 I became ICREA Research Professor. I performed research stays as visiting professor at Stanford University and at UCLA. He has published over a hundred articles in high-ranking international journals with an h-index of 45 and over 8000 citations with 1250 citations per year in 2021. 

Research interests

The group's research interests are rooted in the applications of computation to science, where we regard intelligence as a form of computation itself. 

1) Molecular simulation and machine learning. We use computation such as physics-based simulations and modern machine learning to provide novel, innovative methodological approaches in biomedicine.   

2) Computational intelligence. We investigate machine learning methods that would bring machine intelligence closer to human-level intelligence. We train intelligence agents using reinforcement learning in virtual environments, we built scalable software for reinforcement learning and low-sample learning. 

Selected publications

- P Thölke & G De Fabritiis 2022, 'TorchMD-NET: Equivariant Transformers for Neural Network based Molecular Potentials', International Conference on Learning Representations (ICLR), Conference proceedings.  

- Varela-Rial, Alejandro; Maryanow, Iain; Majewski, Maciej; Doerr, Stefan; Schapin, Nikolai; Jimenez-Luna, Jose; De Fabritiis, Gianni 2022, 'PlayMolecule Glimpse: Understanding Protein-Ligand Property Predictions with Interpretable Neural Networks', Journal Of Chemical Information And Modeling, 62, 2, 225 - 231.

Selected research activities

OpenMM. We are co-PI together with J. Chodera MSKCC USA and T. Markland U. Stanford in the scientific and technical development of OpenMM, one of the leading open-source molecular simulation packages. 

Torch/RL. PyTorch is the most used framework for machine learning. We are key contributors to the upcoming reinforcement learning (RL) library from PyTorch currently in the alpha stage as we had already developed PyTorchRL (https://arxiv.org/abs/2007.02622).