Chemistry Machine Learning
I am conducting an Honors Thesis in Chemistry ML in the Computer Science Department under Professor Jean-Baptiste Tristan.
I am studying geometry optimization of (initially simple) molecules using gaussian process regression and a transferrable acquisition function. We aim to achieve more efficient exploration of the PES using fewer steps.
Our initial system is the optimization of the association of simple hydrocarbons on a doped metallic catalytic surface.
In the future I aim to expand this system to optimize on the latent space of conformational autoencoders (like e3nn or Bayer) to allow the system to scale to larger molecules with more degrees of freedom.
Degradation of Aflatoxin by Secreted Enzymes
Full quantum mechanical modeling of the enzyme-substrate system: how laccase detoxifies aflatoxin. Marco Zaccaria, William Dawson, Darius Russell Kish, Massimo Reverberi, Marek Domin, Luca Dellafiora, Takahito Nakajima, Luigi Genovese, Babak Momeni. bioRxiv 2020.03.02.973883; doi: https://doi.org/10.1101/2020.03.02.973883
I performed homology model preparation, docking studies and post-hoc scoring to show that the level of theory used in molecular mechanics-based docking is not sufficient to capture differences in degradation rates of AFB1 and AFG2.
I am developing a pipeline for identification of secreted enzymes that degrade aflatoxin from a organismal proteome as part of a Spring 2021 research
Consensus in Distributed Systems
Working with Professor Lewis Tseng in spring 2020, I implemented consensus algorithms in custom system environments using MiniNet to benchmark and compare the performance of new algorithms against the current state of the art.
Cassandra+: Trading-Off Consistency, Latency, and Fault-tolerance in Cassandra; Guo-Shu Gao, Kishori Konwar, Juan Mantica, Haochen Pan, Darius Russell Kish. Lewis Tseng, Zezhi Wang, Yingjian Wu; ICDCN 2021 (authors in alphabetical order)
I implemented a number of approximate consensus algorithms in Python and benchmarked with MiniNet. See the code here.
Working in Professor Babak Momeni's lab, I ported a chemical mediator-explicit microbial community model to parallel Julia, leading to an 80x speedup over serial MATLAB.
With this speedup I was able to screen hyperparameters for the model and determined two optimal regions with distinct strategies for coexistence of species.
I presented the results as a solo flash talk at the Boston Bacterial Meeting 2020.
Molecular Dynamics Simulations
Earlier in my academic career I worked on MD simulations for novel cyclic peptides. I developed methodology for performing WHAM free energy analysis in Tinker, a MD package, as well as methodology based on normal mode sampling to tune custom parameters for the CHARMM22 force field in Tinker. Code for NMS-based parameter derivation.