AI-Enabled Uncertainty Simulation in DNA Mechanics

DNA is one of the most fundamental molecules of life. It not only stores genetic information but also behaves like a flexible material that can bend, twist, and loop in specific ways. These mechanical properties depend on the exact order of the four chemical bases—adenine (A), thymine (T), guanine (G), and cytosine (C)—that make up the DNA sequence. Even small changes in this sequence can alter how tightly DNA is packed, how easily it can be read by proteins, and how genes are switched on or off. Understanding how the DNA sequence controls its mechanical behaviour is therefore key to explaining gene regulation and cell organisation.

Research Gap: Although several computer models can simulate DNA’s physical behaviour, most do not measure how confident we can be in their predictions. They often treat DNA as a rigid system, ignoring small variations or uncertainties in experimental data. Meanwhile, artificial intelligence (AI) and machine learning are powerful for identifying patterns in large datasets, but they rarely include the underlying physical laws that govern DNA mechanics. What is currently missing is a combined approach that merges physics-based modelling with AI and data analysis, while also accounting for uncertainty and sensitivity to different inputs.

Aim: This project aims to develop a computational framework that integrates AI and physics-based simulations to predict how DNA’s sequence influences its shape and flexibility. The framework will estimate the confidence level of its predictions and identify which parts of the sequence or model parameters most strongly influence DNA’s mechanical behaviour.

Potential Objectives:

1. Develop sequence-based physical models: Use coarse-grained models to calculate DNA properties—such as bending stiffness, twisting flexibility, and groove geometry—directly from its sequence, while incorporating uncertainty quantification to assess prediction reliability.

2. Integrate AI for predictive modelling: Train machine learning models on simulation and experimental data to make fast, accurate, and large-scale predictions of DNA mechanics.

3. Perform sensitivity analysis: Determine which sequence motifs or parameters most affect DNA shape and flexibility, improving model interpretation and refinement.

4. Validate and benchmark predictions: Compare model outcomes with experimental data, such as DNA looping, nucleosome positioning, and deformation, to ensure accuracy and generalisation.

Overall, the project will develop an AI-enabled, uncertainty-aware simulation tool that predicts DNA behaviour directly from its sequence. This work will establish a next-generation AI platform for molecular biophysics, advancing precision modelling, data-driven discovery, and digital-twin approaches for understanding biological systems.