We work at the interface between computation and the physical sciences, computational intelligence, and computer simulations
Highlights
Over the years we have pioneered several research areas:
Neural network potentials (TorchMD-NET).
Differentiable simulations (TorchMD).
Won several blind challenges in drug discovery and second place in the Unity Obstacle challenge using reinforcement learning.
First and fastest MD on GPUs with ACEMD and co-principal investigator in OpenMM.
First distributed computing projects initially on SONY PlayStation3 and later on GPUs (GPUGRID.net).
Generative models on molecular structures (LigDream, LigaNN, etc).
Machine learning on three-dimensional molecular structures (Kdeep, DeepSite, etc).
High-throughput molecular simulations (HTMD) software.
Bitcoin donations for Science with the project Donate@home.
News
Research
Computer Simulations
The fundamental equations that govern Nature at atomistic scales are well understood in terms of quantum mechanics. Solving such equations is not possible apart from very simple systems, yet solutions to this problem represent one of the grand challenges for computational sciences as it would allow an understanding of all properties of molecular systems. We investigate this challenge by solving the sampling and accuracy problems of atomistic simulations using machine learning, physics, and GPUs.
Computational intelligence
Understanding and replicating human-level intelligence in machines is of great importance for the progress of humanity. We look at intelligence as a physics problem trying to understand the key components. Two key factors in human intelligence are prediction and population learning. Recently, we ranked second in the Unity obstacle tower challenge, an AI testbed for navigation, puzzle solving, etc. We currently own the SOTA on this challenge using PytorchRL.
Projects
Molecular simulations on GPUs ACEMD/OPENMM
OpenMM is a high-performance toolkit for molecular simulation. Use it as an application, a library, or a flexible programming environment. We co-lead OpenMM together with John Chodera from MSKCC, Tom Markland and Peter Eastman from U. Stanford.
A framework for machine learning molecular simulations
TorchMD intends to provide a simple to use API for performing molecular dynamics using PyTorch. This enables researchers to more rapidly do research in force-field development as well as integrate seamlessly neural network potentials into the dynamics, with the simplicity and power of PyTorch.
A modular framework for advancing AI
PyTorchRL is a pytorch-based library for RL that allows to easily assemble RL agents using a set of core reusable and easily extendable sub-modules as building blocks. PyTorchRL permits the definition of distributed training architectures with flexibility and independence of the Agent components.
GPUGRID.net is one of the largest distributed computing project worldwide. GPUGRID pioneered the use of GPUs in distributed computing.
ACEMD has pioneered the use of GPUs for molecular simulations allowing for high-throughput simulations. ACEMD is still one of fastest molecular dynamics code and compatible with input files from Charmm and Amber.
HTMD is a Python platform for computational biology, including molecular simulations, docking, Markov state models, molecule manipulation, build tools, adaptive sampling. Entire experiments can be coded in a single Python script.
Playmolecule is an application-based site that contains methods, predictors and machine learning approaches, most of which are publicly available for drug discovery. Currently managed by Acellera Ltd.
Positions
Apply
We are always looking for talented people who would like to join the laboratory. We praise on diversity of expertises, e.g. mathematics, chemistry, computer science, statistics, physics, biology, biotechnology, etc. If you are a hard-working, self-motivated person with the ambition to do great science, send us an email at gianni.defabritiis@upf.edu.
Publications
Funding
National Institutes of Health
OpenMM is currently (July 2021 - Mar 2025) supported by the National Institute of General Medical Sciences of the National Institutes of Health under award number R01GM140090.
CompBioMed
Computational Methods, based on human biology, are now reaching maturity in the biomedical domain, rendering predictive models of health and disease increasingly relevant to clinical practice by providing a personalized aspect to treatment.
Spin-offs
Developing the algorithms to automate drug discovery.
Contacts
Gianni De Fabritiis, Icrea research professor at University Pompeu Fabra, Head of Computational Science Laboratory, PRBB, Barcelona.
Email: gianni.defabritiis at upf.edu