Michael Gastegger

Michael Gastegger is a machine learning researcher who builds neural networks for chemistry and molecular simulation. Trained as a theoretical chemist at the University of Vienna, he completed a PhD thesis titled Artificial Intelligence in Theoretical Chemistry in 2017, after using machine learning molecular dynamics to predict infrared spectra. His later work includes the Machine Learning Force Fields review in 2020, the PaiNN and SpookyNet networks in 2021, and the cG-SchNet model for designing 3d molecules with TU Berlin's Machine Learning Group in 2022. At Microsoft Research AI for Science, he coauthored BioEmu-1, a deep learning model of protein structural ensembles released in 2025. He is a senior researcher in the lab's sampling team, modelling distributions of biomolecules with machine learning.

Now at: Microsoft
Works: 62
Citations: 6,208
h-index: 24
Badges: Rising
Bridge

Publication numbers from OpenAlex, as of 22 October 2026.

Timeline

2013

Enzyme design in Vienna

He completes a master's thesis on de-novo enzyme design for olefin metathesis in December 2013 in the González Research Group at the University of Vienna's Institute of Theoretical Chemistry. His work there combines quantum chemistry with evolutionary algorithms and machine learning to design artificial enzymes.
2015–2017

Neural networks for infrared spectra

With Philipp Marquetand, he publishes high-dimensional neural network potentials for organic reactions in 2015. In May 2017, he, Jörg Behler and Marquetand use machine learning molecular dynamics to predict infrared spectra, including the first machine learning simulation of a peptide's dynamics.
2017

A PhD and wACSF descriptors

He completes his PhD thesis, Artificial Intelligence in Theoretical Chemistry, at the University of Vienna in September 2017. In December 2017, he introduces weighted atom-centered symmetry functions (wACSFs), descriptors of molecular geometry for machine learning potentials. He also works as a postdoc at the Institute of Theoretical Chemistry.
2020–2021

Machine learning force fields

In October 2020, he coauthors the review Machine Learning Force Fields with Oliver Unke, Stefan Chmiela, Alexandre Tkatchenko, Klaus-Robert Müller and others, including a step-by-step guide to building such models. In May 2021, he, Unke, Chmiela, Müller and others introduce SpookyNet, a deep neural network for force fields with explicit treatment of electronic degrees of freedom and quantum nonlocality.
2021

PaiNN and molecular spectra

In February 2021, he, Kristof Schütt and Oliver Unke propose PaiNN, the polarizable atom interaction neural network, and use it to simulate molecular spectra 4 to 5 orders of magnitude faster than the electronic structure reference. In June 2021, as a researcher at Technische Universität Berlin, he presents his work on machine learning for molecular spectra and solvent effects to the Machine learning in Physics, Chemistry and Materials discussion group in Cambridge.
2022

cG-SchNet and SchNetPack 2.0

In February 2022, as a member of the Machine Learning Group at Technische Universität Berlin and the BASLEARN joint lab of TU Berlin and BASF, he coauthors cG-SchNet in Nature Communications, a conditional generative neural network for the inverse design of 3d molecular structures. In December 2022, he coauthors SchNetPack 2.0, a neural network toolbox for atomistic machine learning with modules for equivariant networks and molecular dynamics.
2025

BioEmu-1 at Microsoft

As a senior researcher at Microsoft Research AI for Science, he coauthors BioEmu-1, a deep learning model of the different structures a protein can adopt, which the lab releases as open source in February 2025. In July 2025, Science publishes the BioEmu paper, describing a system that generates thousands of statistically independent structures per hour on a single GPU.
As of October 2026

Senior researcher in sampling

He is a senior researcher in the sampling team of Microsoft Research AI for Science, where he models distributions of biomolecules with machine learning. In February 2026, he coauthors Enhanced Diffusion Sampling with Frank Noé, Christopher Bishop and others, a framework for rare-event sampling and free-energy calculation with diffusion models such as BioEmu.