
For several decades now there have been steady advances in models and algorithms for nucleic acid structure prediction. These advances have been valuable to biologists who wish to understand the function of RNA molecules in the cell, as well as to molecular programmers and engineers who want to design DNA and RNA molecules with novel functions not found in nature. In contrast, there has been less progress in good models for predicting nucleic acid kinetics - the rate at which molecules fold and unfold, and the structural pathways that they follow during the folding process, although such tools would also be very valuable. In this talk, I'll describe our recent efforts to improve nucleic acid kinetics prediction. I'll first describe our new model of nucleic acid kinetics, based on Arrhenius principles. Unlike previous models, parameters of our model help account for ways in which the rate of nucleic acid base pair formation and breakage can depend on the local context around that base pair. I'll then describe how we calibrate our kinetic model to experimental data, using machine learning techniques. For this purpose, we have curated an experimental dataset with several hundred DNA reactions, including hairpin opening and closing (Bonnet et al.), helix association and disassociation (Morrison and Stols), strand displacement (Reynaldo et al., Zhang and Winfree) and more. Finally, I'll describe how our model fares in predicting nucleic acid kinetics, and how we hope to improve it in the future. This is joint work with Nasim Zolaktaf and Mark Schmidt at UBC, and Frits Dannenberg, Xander Rudelis, Joseph Schaeffer, Chris Thachuk, and Erik Winfree at Caltech.

For several decades now there have been steady advances in models and algorithms for nucleic acid structure prediction. These advances have been valuable to biologists who wish to understand the function of RNA molecules in the cell, as well as to molecular programmers and engineers who want to design DNA and RNA molecules with novel functions not found in nature.

In contrast, there has been less progress in good models for predicting nucleic acid kinetics - the rate at which molecules fold and unfold, and the structural pathways that they follow during the folding process, although such tools would also be very valuable. In this talk, I'll describe our recent efforts to improve nucleic acid kinetics prediction.

I'll first describe our new model of nucleic acid kinetics, based on Arrhenius principles. Unlike previous models, parameters of our model help account for ways in which the rate of nucleic acid base pair formation and breakage can depend on the local context around that base pair. I'll then describe how we calibrate our kinetic model to experimental data, using machine learning techniques. For this purpose, we have curated an experimental dataset with several hundred DNA reactions, including hairpin opening and closing (Bonnet et al.), helix association and disassociation (Morrison and Stols), strand displacement (Reynaldo et al., Zhang and Winfree) and more. Finally, I'll describe how our model fares in predicting nucleic acid kinetics, and how we hope to improve it in the future.

This is joint work with Nasim Zolaktaf and Mark Schmidt at UBC, and Frits Dannenberg, Xander Rudelis, Joseph Schaeffer, Chris Thachuk, and Erik Winfree at Caltech.