Erik Thiede, Ph.D.

I'm an Assistant Professor at Cornell University Department of Chemistry and Chemical Biology, having started Summer 2023. You can find my lab research website here.

Broadly speaking, my research area is theoretical and computational chemistry. This means that theoretically speaking, I can both use computers and do chemistry. I received my PhD working with Profs. Aaron Dinner and Jonathan Weare on enhanced sampling algorithms for molecular dynamics. Currently, I spend most of my time thinking about how we can take biomolecules and biomolecular data, feed them into a computer, and get scientific insight back out. Download my CV.

Research

My research has focused on the intersection of data science, machine learning, and statistical approximation with chemistry. Below are some projects I have worked on in the recent past.

Extracting Free Energies from Cryo-Electron Microscopy

Cryo-electron Microscopy has recently emerged as a leading technique for extracting protein structures. But in principle, Cryo-EM should give us the entire range of possible structures for a given macromolecule, allowing us to extract biological free energies. Working with the Structural Molecular Biophysics Group at Flatiron, I developed new algorithms to extract free energies from Cryo-EM data.

New approaches for learning on permutation-equivariant structures

While symmetry under permutation is fundamental to chemistry, it also restricts the space of possible learning algorithms: if not constructed carefully, algorithms that attempt to learn new structures must face a combinatorial explosion in the number of terms. I am worked with Prof. Risi Kondor to develop new approaches to learning on chemical systems by intelligently exploiting and learning substructures of a chemical system.

Error Estimates for MBAR

The Multistate Bennett Acceptance Ratio is a popular framework for estimating free energies from multiple biased simulations. Historically, the ways in which sampling from the individual biased simulations leads to overall error have been poorly understood. In recent work, Sherry Li and I derived error estimates for the MBAR equations. Our work is the first to rigorously account for the affect of sample correlation and can trace how sampling in individual states affects the total error.

Selected Papers

  1. Li, Xiang Sherry, et al. Understanding the Sources of Error in MBAR through Asymptotic Analysis. arXiv preprint arXiv:2203.01227 (2022). arXiv
  2. Thiede, Erik H., Wenda Zhou, and Risi Kondor. Autobahn: Automorphism-based graph neural nets. Advances in Neural Information Processing Systems 34 (2021): 29922-29934. PDF
  3. Giraldo-Barreto, Julian, et al. A Bayesian approach to extracting free-energy profiles from cryo-electron microscopy experiments. Scientific Reports 11.1 (2021): 1-15. PDF
  4. Thiede, Erik H., et al. Galerkin approximation of dynamical quantities using trajectory data. The Journal of Chemical Physics 150.24 (2019): 244111. PDF

Contact