(2w) Development of Computational Tools for Peptoid Structure-Property Prediction

Peptoids are synthetic biocompatible peptidomimetics that have been used to develop
antimicrobial agents, lung surfactants and drug delivery vehicles. They are protease resistant
and have enhanced cellular uptake, which make them attractive candidates for biological
applications. Since peptoids lack native backbone hydrogens connected to an electronegative
atom and hence have no backbone hydrogen bonding, their secondary structure is governed
primarily by steric interactions. Based on the sidechain type and sequence, residue specific
interactions can promote or inhibit formation of secondary structures. Furthermore, unlike
peptides, which have trans configuration as the prevalent amide bond isomer, peptoid amide
bonds can have both cis- and trans- configurations. This allows peptoids to exhibit a variety of
secondary structures that are not observed in peptides.

Due to these differences, forcefields developed for peptides have very limited applicability for
peptoids. Using CGenFF-NTOID, an atomistic peptoid forcefield developed in the Santiso lab,
we were able to extend the forcefield to multiple sidechains. We are able to obtain correct global
and local minima predictions through enhanced sampling simulations. We also use these newly
developed parameters to study secondary structure formation by peptoid oligomers where we
see É‘-helix structure formation along with the free energy barriers for transitions from cis- to
trans- forms. This forcefield provides an important step for accurate structure prediction of
peptoid oligomers.

We also developed a coarse graining methodology using the CGenFF-NTOID atomistic
forcefield. Our technique uses Relative Entropy bottom-up coarse graining along with SAFT-γ-
SW equation of state to develop parameters for atoms grouped together as beads. We have
been able to approach multi-millisecond peptoid simulations through this coarse graining
implemented in an in-house Discontinuous Molecular Dynamics (DMD) architecture. This
framework is one of the first of its kind, and our work forms the basis for a combinatorial
peptidomimetic library for materials discovery applications.

In order to further analyze peptoid solution behavior, we are also working on a machine learning
(ML) based small molecule solubility paradigm for peptoids in organic solvents. Due to the
absence of solubility data for peptoids, our model is trained on a wide variety of organic
molecule solubilities. To enable our model to make temperature-based predictions where
experimental data is scarce, we also incorporate a thermodynamics equation of state to our ML
technique.

Research Interests

Through my research, I have gained experience in a wide variety of simulation methods, such
as conventional and discontinuous Molecular Dynamics, enhanced sampling methods, Monte
Carlo and ab-initio techniques. I have also employed tools such as coarse graining and
supervised learning techniques along with neural networks to study biomolecular systems. In
my doctoral degree, I have obtained programming and software development skills across a
variety of languages (C++, Python, R). In my postdoctoral research, I want to expand my
knowledge and develop different kinds of machine learning algorithms, along with their
applications to systems of materials and biological importance. In my future research, I hope to
utilize my skills to develop methods which enhance our understanding of biomaterials for
potential applications as therapeutics and novel materials discovery.