Biomolecular Modelling and Design

Biomolecular modelling and design integrate theoretical and computational methods to predict, analyse and engineer the structure, dynamics and function of biological macromolecules. At its core, molecular mechanics and molecular dynamics simulations employ empirical force fields to explore conformational landscapes of proteins, nucleic acids and their complexes, revealing mechanisms of folding, binding and catalysis. Quantum-mechanical approaches, from density functional theory to wavefunction-based methods, yield atomistic insight into reaction pathways and electronic structure. In recent years, machine-learning potentials and generative algorithms have closed the gap between accuracy and scale, enabling near-ab initio precision in simulations of large biomolecular assemblies. These advances support rational design of novel enzymes, therapeutic antibodies and de novo folds, underpinning breakthroughs in biocatalysis, drug discovery and synthetic biology. By combining high-throughput experimental data with predictive models, biomolecular design has become a cornerstone of efforts to address global challenges in health, sustainability and biotechnology.

Research from Nature Portfolio

A massively parallel experimental framework now quantifies folding stability for nearly one million protein domains in a single workflow. By linking cDNA display with proteolytic selection, researchers have mapped how every single-residue variant alters thermodynamic stability across hundreds of natural and designed mini-proteins. This “atlas” of folding energetics exposes long-range couplings between distant sites and refines the principles that guide scaffold engineering.

A universal graph neural network potential, pretrained on millions of density functional theory trajectories, explicitly incorporates atomic charges and magnetic moments. It achieves accurate molecular dynamics of complex inorganic and organic systems, from battery cathodes to enzymes, capturing ionic and electronic degrees of freedom without repeated quantum calculations.

An E(3)-equivariant neural network demonstrates that embedding rotational and translational symmetries yields highly data-efficient interatomic potentials. With orders-of-magnitude fewer training points, it reproduces quantum forces for diverse molecules and materials, proving that symmetry-aware architectures can underpin predictive biomolecular simulations at scale.

Research from all publishers

Hybrid enzyme design frameworks now merge deep-learning structure predictions with multiscale simulations to propose site-specific mutations that enhance catalytic performance. Generative models trained on sequence-structure databases navigate vast sequence space, yielding biocatalysts with improved activity, stability and solvent tolerance for industrial processes.

Artificial-intelligence-driven de novo protein design has matured to produce entirely new folds and programmable functions. By integrating physics-based modelling with generative networks, researchers have created modular proteins capable of precise molecular recognition, self-assembly and synthetic signalling, opening frontiers in synthetic biology and therapeutic engineering.

Biomolecular Modelling and Design publication trend

The graph below shows the total number of articles in biomolecular modelling and design across all publications each year (not limited to Nature Index journals).

Technical terms

Molecular dynamics simulation: Time-resolved computational method that computes atomic trajectories by integrating Newton’s equations of motion under a defined force field.

Force field: Empirical functional form and parameter set that approximates potential energy surfaces of biomolecular systems.

Machine-learning potential: Data-driven model trained on quantum calculations to predict energies and forces for large-scale atomistic simulations.

E(3)-equivariance: Property of a model ensuring outputs transform correctly under three-dimensional rotations, translations and reflections.

Proteolytic selection: High-throughput experimental technique that uses selective cleavage to assess folding stability of protein variants en masse.

De novo protein design: Computational strategy to generate entirely new amino-acid sequences predicted to fold into specified three-dimensional structures.

References

1. Mega-scale experimental analysis of protein folding stability in biology and design. Nature (2023).
2. CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling. Nature Machine Intelligence (2023).
3. E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials. Nature Communications (2022).
4. Navigating the landscape of enzyme design: from molecular simulations to machine learning. Chemical Society Reviews (2024).
5. De novo protein design—From new structures to programmable functions. Cell (2024).