Computation applied to Science

The group's research interests are rooted in the applications of computation to science, in particular to physical simulation and understanding intelligence seen as a form of computation.

Highlights

PlayMolecule

PlayMolecule is a platform for drug discovery and structural biology applications based on physics-based simulations and deep learning. Try it out here.

Unity Obstacle Tower RL challenge

We classified second at the Unity obstacle tower challenge, where an agent had to navigate a complex 3D environment and solve tasks. Here is the agent in action: Video, and the Unity blog post here.

LogP SAMPL 2019 challenge

We scored second to the logP small molecule blind prediction challenge.

D3R 2018 challenge

We won two sub-challenges in the D3R 2018 drug discovery blind challenge in collaboration with Acellera related to binding affinity and pose prediction of protein-ligand complexes.

Research

Research lines

Computer Simulations

Physics-based simulations and modern machine learning methods provide innovative methodological approaches in biomedicine which can be applied to drug design, protein folding, protein-protein interactions, etc. We created GPUGRID.net, the second largest distributed computing project harnessing several thousands GPUs. We are core project leaders of OpenMM/ACEMD, one of the leading molecular dynamics packages. We created PlayMolecule, a publicly-available platform offering molecular simulations and machine-learning-powered assets for drug discovery used by a wide community of users.

Machine Intelligence

Understanding and replicating human level intelligence in machines is of great importance for the progress of humanity. We look at realistic 3D simulated environments for navigation and robotic manipulation to develop learning methods which can be transferred to the real world. We are particularly interested in solving hard robotic challenges by training intelligent agents using super-scalable population learning methods using GPUGRID and low sample learning (arXiv:2007.03328) via imitation, self-imitation and model-based. 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 (arXiv:2007.02622)

Software

HTMD

HTMD is a Python platform for computational biology, including molecular simulations, docking, Markov state models, molecule manipulation, build tools for Amber and Charmm, visualization (webGL and VMD), adaptive sampling and more. Imagine setting up an entire computational experiment in a single, simple Python script.

ACEMD/OpenMM

ACEMD/OpenMM has pioneered the use of GPUs for molecular simulations allowing for high-throughput simulations and ultimately leading to HTMD. ACEMD is still one of fastest molecular dynamics code and compatible with input files from Charmm and Amber. ACEMD is now based on OpenMM.

PyTorchRL

PyTorchRL is a scalable and modular reinforcement learning framework in PyTorch. Use it to write new algorithms and scale them up for real performance. This code currently hold the SOTA on Obstacle tower challenge by Unity3D.

OpenMM

OpenMM is a high-performance toolkit for molecular simulation.

TorchMD

TorchMD - A deep learning framework for molecular simulations.

Online resources

PlayMolecule

Playmolecule is an application-based site that contains methods, predictors and machine learning approaches, most of which are publically available for drug discovery. Currently managed by Acellera.com.

GitHub

The Computational Science Laboratory GitHub page contains more tools. Follow us there.

GPUGRID.net

GPUGRID.net is one of the largest distributed computing project worldwide. GPUGRID pioneered the use of GPUs in distributed computing.

Apply

We are always looking for talented people who would like to join our 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.

Closed Positions

If you are interested in doing a specific project idea with us send a CV and motivation letter to gianni.defabritiis@upf.edu

Contact