Understanding protein dynamics
Proteins are so small that they cannot be directly observed. Furthermore, protein dynamics occur too quickly to perceive, even if we could observe them. Experimental methods for monitoring protein motion often produce very coarse representations of just a few moving parts. Thus, many researchers have sought to model protein dynamics using computer simulations, which enable researchers to "see" every atom in a protein move at very short (picosecond!) timescales.
In order to make robust conclusions, it is necessary to interpret such simulations in a statistical manner, instead of simply watching what happens. My PhD research focuses on building statistically rigorous protein dynamics simulations in order to model and understand what a protein is doing when it moves.
The type of model I use is called a Markov state model, which assigns to a state every frame in the simulation (like frames in a movie reel). We typically have millions of frames and automatically assign each frame to one of a few hundred states. Thus, we can analyse states in terms of their populations, the probabilities of transitioning between them and the possible pathways from any state to any other state. These quantities correspond to scientific variables of interest such as energy, kinetic rates and transition pathways.
By creating Markov state models for simulations of protein dynamics, we can better understand biophysical processes, for example, (1) how proteins fold and misfold, the latter being implicated in various diseases; (2) how folded proteins change conformation, which might reveal new stable structures that we can propose small molecule drugs to stabilise or destabilise; and (3) how and where small molecules bind and unbind to a protein, which yields information about how drugs interact with the protein to which they bind.
While many of my colleagues focus on modeling specific systems with medical applications, such as opioid receptors for pain management applications and kinases for cancer research, my PhD research centres on developing the models themselves from a mathematical viewpoint in order to enhance the research of my colleagues. For example, I have recently developed an algorithm to address the challenging problem of reducing a hundred-state model to a two- or three-state model, which is useful for interpreting models as well as recognising what is most important in the system.
Over the past two decades since Markov state models were proposed for studying the dynamics of molecules, there have been many mathematical improvements to the method, but there was no single place that all 20 years of advances were presented for a general scientific audience. Therefore, after reading about 400 journal publications, I recently wrote a perspective review for The Journal of the American Chemical Society summarising the developments of the method. My aim was to explain and promote the growth of the subfield and to present the most recent discoveries it has enabled as well as future directions researchers should pursue.
Ultimately, one of my goals as a scientist is to communicate complex ideas in an accessible way that enables non-experts to adapt the strategies and algorithms that many researchers (including myself) have worked so hard to develop. By developing algorithms designed with interpretability in mind, and by publishing perspective articles communicating new methods for a general scientific audience, I seek to make progress towards this goal.
*Brooke Husic did an MPhil in Chemistry at the University of Cambridge and is now pursuing a PhD at Stanford.