Methodological Innovations in Computational Biology

Turning molecular trajectories into mechanistic explanations

A molecular simulation is useful when it explains a biological observation or makes a testable prediction. The challenge is to sample the relevant conformations and identify the interactions that govern transitions between them. We develop and apply computational methods that connect local atomic behavior to collective protein motion, molecular recognition and function.

Our work combines all-atom molecular dynamics with elastic network models, enhanced sampling and free-energy calculations. The level of resolution follows the question: coarse-grained models describe cooperative motions across a protein, while atomistic simulations resolve the interactions, solvent environment and conformational changes needed to explain a particular mechanism.

Multiscale sampling of conformational transitions

Our collective molecular dynamics method, coMD, couples elastic network descriptions of global motion to atomistic molecular dynamics. The published adenylate kinase application showed how this combination can explore large conformational changes while retaining atomic detail. Our related work on a two-state anisotropic network model provided another route for investigating transitions between known conformations.

These methods address the mismatch between the motions accessible to a short conventional trajectory and the larger transitions required for function. We apply this perspective to systems such as dynein and viral spike proteins, where local interactions participate in coordinated structural rearrangements.

Energetics and computational force spectroscopy

Conformational transitions and molecular dissociation depend on energy landscapes, not only on structural differences. We use enhanced sampling and free-energy calculations to identify accessible states and the barriers separating them. In our dynein priming-stroke study, these calculations connected linker motion to the mechanical constraints imposed by the motor ring.

Steered molecular dynamics provides a complementary way to examine resistance to a specified mechanical perturbation. We use it to compare dissociation pathways and force-dependent responses under controlled simulation conditions. Pulling direction and loading rate matter, so simulated rupture forces are interpreted within the protocol rather than treated as interchangeable with equilibrium binding affinity.

Mapping attractive and repulsive interactions

Our 2025 covariance-based method identifies candidate interactions from correlated residue motions, filters them by spatial proximity and classifies the underlying physicochemical interactions. This focuses analysis on relevant parts of a protein-protein interface and captures repulsive interactions that are easily missed by workflows centered on favorable binding alone.

We demonstrated the method with nanobody-spike complexes. Covariance is used together with molecular geometry and interaction analysis; the sign of a correlation is not, by itself, a universal label for attraction or repulsion. When an interface occupies different binding states, analyzing those states separately helps preserve the mechanistic information.

From correlated motion to an interpretable interaction map. The workflow combines an interprotein covariance matrix, spatial filtering, interaction classification and molecular visualization for H11-H4 bound to the Omicron spike receptor-binding domain. Reproduced unchanged from Golcuk and Gur (2025), Figure 1. Golcuk and Gur (2025)

Finding control points in protein dynamics

Elastic network models also help identify global hinge sites that coordinate collective motion. Our 2024 analysis compared these mechanically important regions with experimentally resolved drug-binding sites and found a statistically supported overlap. This provides a physical rationale for exploring conformational control points as possible sites of intervention, connecting method development to therapeutic questions.

Testing the relationship between global hinge sites and drug-binding sites. The workflow compares experimentally resolved drug-bound structures with collective motions predicted by elastic network models. Reproduced unchanged from Zhang, Gur and Bahar (2024), Figure 1. Zhang

A common toolkit across biological systems

The same physical framework supports our studies of motors, antigen presentation, repeat-protein scaffolds, viral recognition and amyloid assembly. Earlier work on neurotransmitter:sodium symporters examined how oligomerization changes transport-related motions. Across these systems, the goal is to connect a defined molecular perturbation to a change in conformational behavior and then to the biological process it influences.

Explore all research domains