Project Portfolio
Examol
Formerly Ibeks Technologies.
Cofounder and CTO.
Working on a cloud-native Kubernetes application for computational drug-discovery.
Designed and implemented our entire server backend in fully statically typed Python code using the Starlette web server.
Roivant Sciences
I work closely with computational scientists on drug discovery teams to develop simulation and ML pipelines. This includes bespoke solutions and general libraries for performing highly scalable analysis platforms tuned for massive simulation data.
In addition I have worked on various aspects of DevOps, workflow orchestration, and data engineering in a a cloud-HPC hybrid organization.
This has included extensive experience with package management, deployment, and compilation of complex sets of dependencies via the excellent Spack package manager.
PhD Related
Thesis Project: Drug Unbinding Transition-State Plasticity
My main thesis project is to investigate the plasticity of ligand unbinding transition states from receptor binding sites.
This is of interest to drug designers which are looking to tailor the kinetic properties of drugs. Such kinetics oriented design can have impacts on drug efficacy and practical improvements of drugs by decreasing the frequency of dosing and avoiding side effects by improving windows of efficacy.
This project pipeline is multipart.
First, we perform molecular dynamics (MD) simulations of drug molecules unbinding from their protein targets using supercomputer clusters of graphics processing units (GPUs).
Because the unbinding process is a slow one (relative to the timescale of the MD simulations) novel enhanced sampling algorithms must be used in order to speed up the simulations without disturbing the physical correctness of the structures we observe.
Primarily this is the WExplore algorithm which was implemented in my
weighted ensemble simulation library wepy.
These methods are similar to Markov State Monte Carlo (MCMC), particle methods, and other importance sampling techniques and are applicable outside of biophysics for estimating probabilities of rare events.
Second, a number of machine learning and custom data mining methods are applied to extract meaning from the "raw" simulation data in a distributed fashion.
This includes geometric constraint queries using the mastic and geomm
libaries, as well as custom analyses with numpy and `scipy``.
I also perform machine learning analyses using sckit-learn for
clustering and PCA as well as a number of libraries specific to the
field for building Markov State Models and calculating stochastic
network properties via Transition Path Theory.
I have leveraged the distributed computing framework dask to scale the analysis of terabytes of simulation data.
The results of this research so far have been published or are still in progress.
Results
Thesis: - Thesis
Data: - zenodo
Posters and Conferences: - Unbiased Molecular Dynamics of 11 min Timescale Drug Unbinding Reveals Transition State Stabilizing Residues