Dr. Frank is developing new tools to build computer models of RNA, DNA, proteins, and the interactions between these various biomolecules.

After moving to the US from Grenada, Dr. Frank completed his undergraduate degree at Brooklyn College. He began his Ph.D. at the University of Michigan and then followed his advisor to UC Irvine. He was supported during his Ph.D. by a predoctoral fellowship from the National Science Foundation. After earning his Ph.D., he worked at a biotech start-up for two years, and then completed post-doctoral work at the University of Michigan (supported by a UM Presidential Post-Doctoral Fellowship). He joined the faculty shortly after.

Much of the work of a cell is carried out by proteins. Protein structure is described at four levels:

1. The DNA sequence of a gene dictates a particular order of amino acids, which are strung together to create the primary structure of a protein.
2. This chain of amino acids (also called a polypeptide) then twists around itself, forming the secondary structure.
3. These local structures are then folded together to form the tertiary structure.
4. Additionally, many proteins are functional only when multiple subunits (i.e., separate polypeptides) come together in a particular arrangement called quaternary structure.

The function of a protein is often deeply intertwined with its structure, particularly folds and grooves created by tertiary and quaternary structure in which other molecules can interact with it. Additionally, many proteins contain unstructured regions, which also have important functions. It is not simple to determine how a protein is structured: biochemists use multiple techniques to learn about protein structure and computer programs help to synthesize these observations into a cohesive 3D model. As we improve these models, we also improve our ability to predict protein structure based on sequence alone.

In a recent paper, Dr. Frank and his former postdoctoral advisor describe a new computer model for predicting how a protein will interact with different solvents (things it is dissolved in). They examine that their model has two main advantages over previous methods to do the same thing. First, it has a lower computational load — it can generate equally good models while doing fewer/easier calculations, which will allow future models to repurpose some of the computational power toward something else. Second, it is less biased toward complicated structures — many proteins have distinct structured and unstructured forms which play different biological roles. Unlike previous models, Dr. Frank’s model is equally good at predicting more random (unfolded) proteins as it is at predicting highly-structured proteins.