Vincent Voelz
LightNews LightNews
voelzlab.bsky.social
Vincent Voelz
@voelzlab.bsky.social
190 followers 230 following 30 posts
Professor of Chemistry at Temple University. Molecular simulation, stat mech, ML/DL, protein dynamics, biophysics, Bayesian inference, computational design, drug discovery.

https://orcid.org/0000-0002-1054-2124
https://sites.temple.edu/voelzlab
Posts Replies Media Videos
voelzlab.bsky.social
To demonstrate the power of our approach, we used BICePs to reweight MSMs of chignolin simulated in nine different force fields, and calculated the BICePs score to rank them. For this system A99SBnmr1-ildn is the most consistent with the combined data.
Ranking of force fields for chignolin CLN001 using the total BICePs score calculated using various likelihoods and different types of experimental observables (500 microstates). The smaller the score, the greater the evidence supporting the model. Each column reports values relative to the minimum value.
June 29, 2025 at 3:46 PM
voelzlab.bsky.social
As an application, we built an MSM of the miniprotein chignolin using the A99SB-ildn force field, and then applied BICePs to reweight populations using a set of 158 NMR measurements. Even though the force field favors the misfolded state, BICePs is able to correctly reweight the landscape.
Reweighted macrostate populations of chignolin CLN001 for various data models. (a) Representative structures of macrostates derived from a 500-state MSM built from ∼30 μs simulations in A99SB- ildn. (b) Prior MSM macrostate populations and BICePs-reweighted populations for five different models are compared against experimental folded-state populations of∼61% from Honda et al. (we assume the remaining population comprises unfolded and misfolded populations).
June 29, 2025 at 3:46 PM
voelzlab.bsky.social
What is the BICePs score? It’s the free energy of “turning on” the experimental restraints and the predicted populations. The better these two agree, the lower the free energy will be.
A graph schematically showing how the BICePs score is calculated. Conformational population distributions are "turned on" using a parameter ξ=0→1, and the restraint potential is turned on using a parameter λ=0→1. The free energy of the transformation is the BICePs score. Shown are simple 1D Gaussian distrinbutions for two models: Model 1 has good overlap between the Bayesian prior (theory) and the likelihood (experimental restraints), while Model 2 has less overlap. The computed BICePs score for Model 1 is thus lower than the score for Model 2.
June 29, 2025 at 3:46 PM
voelzlab.bsky.social
Does it work in realistic systems? You bet! Here’s one leg of a relative binding free energy calculation where we alchemically transform Ala to Phe using #pmx in #GROMACS. Optimized intermediates yield EE simulations with much better convergence.

5/7
(A) A Visualization of the A8 miniprotein rendered using UCSF ChimeraX, with the protein backbone as ribbons, and the A16F residue involved in alchemical transformation shown as "licorice". (B) A comparison of the 45 unoptimized intermediates (top) 26 optimized intermediates.(C) Traces of average free energy estimates over time for five 40-ns EE simulations using the unoptimized values (black), and five 40-ns EE simulations using the optimized values (green). Shaded regions denote standard deviations over the five trajectories. (D) Traces of the average Wang-Landau bias increment over time for unoptimized (black) versus optimized (green) simulations. Shaded regions denote standard deviations.
June 23, 2025 at 4:10 PM
voelzlab.bsky.social
More mixing leads to smaller uncertainties. Sure enough, EE sampling with the optimal number of intermediates (K*=25) gives the lowest uncertainty.

4/7
A graph of the error of the free energy estimate ∆f (in units k_B T) versus the number of intermediates (K) show that the optimal number of intermediates, K*=25, leads to the smallest error.
June 23, 2025 at 4:10 PM
voelzlab.bsky.social
Using the same approach, we optimize the number of intermediates used in EE simulations. Modeling the exchange acceptance as Markov process reveals the optimal number of intermediates that minimize the mixing time.

Here’s a result for a toy system of 1D harmonic potentials:

3/7
A graph of the computed mixing time (t_2) vs. the number of intermediates (K) shows that using K=25 intermediates leads to the fastest mixing time.
June 23, 2025 at 4:10 PM
voelzlab.bsky.social
First, we estimate the thermodynamic length between intermediates using a short expanded-ensemble (EE) simulation (available in @gromacs.bsky.social thanks to @michaelshirts.bsky.social ), and then use a cubic spline fit to do gradient-based optimization of the spacing.

2/7
A graph of the thermodynamic length L vs. alpha illustrates how alchemical intermediates alpha_i can be adjusted to equalize the thermodynamic length between intermediates.
June 23, 2025 at 4:10 PM
voelzlab.bsky.social
Spent the day in D.C. at the #standupforscience2025 rally! Huge props to organizers and everyone at @standupforscience.bsky.social for a great event. Stand Up For Science!
March 8, 2025 at 2:47 AM
voelzlab.bsky.social
What to my wondering eyes should appear, but UCSF BMI student and Temple alum Rashad Reid visiting campus! #TempleMade @templechemistry.bsky.social @temple.edu
Temple alum Rashad Reid (left) posing for a selfie with Chemistry Professor Vincent Voelz (right)
December 17, 2024 at 4:06 AM
voelzlab.bsky.social
Great talk from first-year Bioinformatics PhD student Jason Cargill at the Temple BIO-BINF Rotation Talks!
Bioinformatics PhD student Jason Cargill presents a slide titled “Exploring conformational dynamics of Pin1 through molecular dynamic simulations”
December 10, 2024 at 3:50 PM
voelzlab.bsky.social
Jumping on the word cloud trend. scholargoggler.com
Cloud of words in various colors and sizes based on the Google Scholar profile for Vincent Voelz. Some of the largest words include molecular folding models markov binding using state simulation dynamics protein simulations model computational conformational kinetic peptoid experimental approach bayesian network inference mutation sars-cov-2 cyclic free mechanism peptide biceps data discovery
November 24, 2024 at 2:53 PM