Center for Biophysics and Quantitative Biology

The notion of energy Landscapes informs work ranging from cosmology to neurology. The theory of energy landscapes has been most quantitatively developed in the “middle realm” from materials to biomolecules. I will first describe how the energy landscapes of glasses are now so well understood as to provide amazingly quantitative predictions.

The energy landscape theory of protein folding and machine learning/artificial intelligence have a long common history.

Starting in the late 1980’s energy landscape ideas suggested that efficient folding required the selection of funneled landscapes. This insight powered the design of structure prediction codes using bioinformatic input and the success of such codes has grown over the decades. A decade ago a great leap in accuracy was achieved when the explosion of sequencing efforts enabled the efficient use of co-evolutionary analysis in machine learning. The physical and evolutionary energy landscapes are correlated. Evolutionary selection however requires not only funneled folding landscapes but also frustrated parts of the landscape that allow functional binding to targets and allosteric motions. This functional frustration shows up in the co-evolutionary analysis and improves structural predictions over what can be done based on a single sequence.

I will discuss how the analysis of frustration gives insights both into the evolution and the de-evolution of proteins and the exon-intron problem. Frustration analysis can also empower machine learning tools such as Alphafold to not just predict static structures but also to uncover functional pathways of protein motion. Energy landscape ideas also provide new tools for drug design.