From predicted structures to predicted ensembles: how artificial intelligence is reshaping molecular modeling in nucleic acids research
Nucleic Acid Insights 2026; 3(9), 625–632
DOI: 10.18609/nai.2026.082
Artificial intelligence is rapidly transforming nucleic acid research and many other scientific disciplines. In molecular modeling, AI is poised to simplify system preparation and accelerate computational workflows rather than replace physics-based molecular dynamics simulations. AI tools will work alongside computational chemists to identify collective variables, extracting information from trajectories, generating structural hypotheses and prioritizing which molecules should be synthesized and tested. This convergence is relevant for antisense oligonucleotides (ASOs), siRNAs, RNA-targeted ligands, and other chemically modified nucleic acids, where activity depends on conformational ensembles, solvation, ion distributions, hybridization thermodynamics, protein recognition, and modification-specific chemistry. Recent developments in AMBER, CHARMM and GROMACS illustrate this shift. These platforms now provide increasingly direct interfaces to machine-learning potentials, multiscale ML/MM calculations and GPU-accelerated sampling. The emerging opportunity is therefore a hybrid one: AI can propose and prioritize, whereas molecular dynamics can test physical plausibility, quantify ensembles and energetics, and expose failure modes. For nucleic-acid drug discovery, the most useful future is likely a closed loop in which experiment, AI, and molecular simulation continuously inform one another.