Hello World!
Thanks for taking the time to look at my portfolio.
For scientific work, have a look at my CV. For programming and data science projects, see below.
Our lab has been working with a number of collaborators on a study of lateral flow assays for the detection of SARS-CoV-2 antibodies. We've designed and implemented an interactive app to complement the study. Interested parties getting an antibody test are advised to speak with their healthcare provider about interpreting their results with that app, which uses local COVID19 prevelance data from usafacts.org (also used by the CDC) and measurments from the study to estimate the probability they do or do not have SARS-CoV-2 antibodies given their result.
ContrApption: An R package providing interactive JavaScript visualizations for RNA-Seq style datasets
Check it out here.
My first foray into "hacktivism" involved working on this web app.
A misleading error message and some weird factor behavior is all it takes to spoof the results of a differential expression analysis. You have been warned!
This is based on an observation I made while exploring CDC gun deaths for machine learning practice.
An R Shiny app that permits interactive visualization of the Wisconsin breast cancer dataset. There is also an extra, fictitious (randomly generated) "cohort" variable to permit demonstration of some additional features (tracking members of the same cohort across multiple graphs). Check it out the video.
ASH (Antigen Selection Heuristic) is a working prototype bioinformatics tool that finds chemically distinct regions in proteins. When designing an antibody to target a protein, it is often beneficial to know which regions are the most distinct. This helps manage the odds of cross-reactivity. For example, if you want to study XYZprotein1 but not XYZprotein2, selecting a region that composed of the same amino acid sequence in both of them will make it likely that your assays are sensitive to both targets.
ASH uses a scale based on hyrdophilicity to compare residues. It used this scale to generate a "mismatch" score to quantify how unlike one another regions are based on their chemical traits, not just a match/no-match system. There is a Jupyter notebook in the ASH directory detailing the algorithm and showcasing prototype Python code. There is also a subdirectory called "ASH++" that shows how the algorithm could be re-written in C++ for greater performance. There is a walkthrough of the basic idea, and the full implementation with documentation and unit tests is here.
Because analyzing text presents numerous challenges in the form of word-stemming, context, and tone, I wrote a tweet classifier that uses only numeric metadata to predict determine if a given message came from the handle of Hillary Clinton or Donald Trump. It uses a Kaggle dataset that I heavily re-engineered using Python scripts to more than double its wide. The classifier() works with ~97% accuracy.
Here you will find smaller programs where practice a variety of languages: Python, Perl, C, and Java. For R code, see the above classifier as I generally use it for Data Science tasks only.
Sometimes you need to have a little stochastic fun and write a Monte Carlo Simulation of the Monte Hall Paradox.
Thanks again for reading, and feel free to reach out with any questions through the contact link above.
Regards: ----Thadryan