A higher accuracy in modelling small molecules bound to proteins has been achieved by switching how candidate structures are chosen. Previously, this selection relied on strict limits; now, it’s reformulated as an optimisation problem solved using a quantum, classical hybrid solver called Quantum-qFit.

In over seventy-four per cent of cases, Quantum-qFit produced better agreement with experimental data and reduced structural strain compared to existing methods. An improved computational method refines how scientists select different shapes a molecule can take when interacting with proteins; this is important because molecules don’t always adopt one fixed shape within a protein’s binding site.

Models more closely matched experimental data in over seventy four per cent of cases and exhibited reduced structural strain by reframing selection as a mathematical problem solved using quantum computing alongside conventional methods. The team at Yonsei University has refined computational methods used to build more accurate models of how small molecules interact with proteins, recognising that these molecules rarely adopt just one shape within a protein’s binding site.

Their approach relied on strict rules, akin to choosing the most efficient route on a map with multiple constraints, when selecting potential molecular shapes; now they have recast this as an optimisation problem solved using quantum computing alongside conventional techniques. This shift in formulation alters which ligand states are represented by available data but does it translate into stronger support at the atomic level.

Quantum Computing Enables More Accurate Protein Ligand Conformations

A substantial improvement across multiple metrics was achieved with Quantum-qFit: higher ligand-region real-space correlation coefficients (RSCCs) were delivered in 62 out of 83 cases (74.7%) alongside lower occupancy-weighted torsional strain in 66 entries (79.5%). Previously, achieving accuracy when modelling small molecules bound to proteins faced limitations due to strict conformation selection boundaries that hindered exploration beyond pre-defined shapes.

By reformulating this process as an optimisation problem solved using quantum computing and conventional techniques, researchers created a framework capable of identifying more accurate molecular shapes within protein binding sites. The new workflow offers flexibility not found in traditional mixed-integer quadratic programming approaches by employing penalties rather than absolute constraints during conformer selection.

Quantum-qFit demonstrated reduced occupancy-weighted torsional strain, a measure of how naturally a molecule bends within its environment, across 66 out of 83 tested cases, indicating more realistic binding poses. Analysis of 28 instances where both modelling approaches arrived at the same number of preferred shapes revealed consistent directional trends in results; these shifts weren’t random fluctuations but systematic changes in conformational preference.

While higher regional map agreement was observed in 62 entries (74.7% with improved RSCC scores), molecular Electron Density for Individual Atoms decreased in most cases, highlighting an important subtlety: better overall fit doesn’t always translate to stronger density support at each atom.

Optimising Ligand Conformations with Classical and Quantum Computation

Accurately modelling how small molecules bind to proteins demands accounting for their flexibility, their ability to adopt multiple shapes within a binding site. Determining which conformations are truly relevant remains challenging, however; Yonsei University’s team has demonstrated an alternative approach by reframing this conformer selection as a mathematical optimisation problem solved using both conventional and emerging quantum computing techniques. This new workflow alters ligand conformations appearing in final models derived from experimental data without changing the underlying inputs, achieving improved accuracy in over seventy-nine per cent of cases when compared with existing methods.

Even acknowledging that assessing true relevance amongst many possible molecular shapes is inherently difficult, this work represents major progress in how scientists model ligand binding. Reframing conformer selection, the process of choosing which shape a molecule takes when attaching to a protein, as an optimisation problem allows both traditional and quantum computers at Yonsei University to tackle it more efficiently; accurately modelling flexibility is key. The approach could unlock more effective drug design and accelerate structural biology research within the coming decade.

Accurately determining the preferred conformations of molecules bound to proteins is essential for understanding biological processes. Researchers reformulated the method used to select these relevant conformations as a mathematical problem solvable with both conventional and quantum computing techniques. While improved regional map agreement did not always equate to stronger atom-level density support, results demonstrate systematic changes in conformational preference compared to existing methods.

👉 More information
🗞 QUBO-Based Conformer Selection with Quantum-Classical Hybrid Optimization for Multiconformer Ligand Modeling
✍️ Jiyun Lee, Haeseong Kim, Junho Lee and Joonsuk Huh (Yonsei University)
🧠 ArXiv: https://arxiv.org/abs/2610.29058

More like this