Doggy Database is a full stack web application designed to manage dog adoptions, shelter operations, and reporting for an animal shelter. The system is built using a modern microservices architecture, leveraging Docker for containerization, PostgreSQL for data storage, Flask for backend APIs, and React for the frontend user interface.

Github: Dog Shelter Database

Analyzed deep learning models (EternaFold, MXFold2) for RNA structure prediction across varying sequence lengths and motifs. Utilized Mxfold2 to identify new sequence liabilities for RNA constructs.

Improved model robustness through hyperparameter tuning and data augmentation. Implemented super ensemble methods combining predictions from multiple models using weighted voting.

Github: Deep Learning for RNA Secondary Structure Prediction

Developed an appointment reservation application that runs on the command line terminal and connects to a Microsoft Azure database server to help caregivers organize vaccinations. Wrote SQL commands and Python code within VS Code to update a server database via pymssql. Kept track of inventory, usernames, logins/logouts, appointments, and securely stored passwords by salting and hashing. Coded relational databases, transactions & commits, two-phase locking, and key management.

Github: Vaccination Database

The protein function of a significant part of sequenced genomes are unknown. This is due to the high variability in secondary structures and functions that can result from the same subsequence. Consequently, a computational method for estimating secondary based on sequence provides a potent tool in resolving the structure and function of yet uncharacterized protein sequences. Primary protein sequence data was curated from the Protein Data Bank and mapped to secondary structure data generated by the DSSP. Sklearn was utilized to generate a decision tree classifier model on encoded categorical data compiled and cleaned from Kaggle. The model was tuned with hyperparameters on frame size and tree depth. Statistical analysis was performed on error and data was visualized with seaborn.

GitHub: Machine Learning Methods to Predict Protein Secondary Structure

We report a novel, three-dimensional bone tissue model as a platform for musculoskeletal disease modeling that allows for compressive loading. By seeding induced pluripotent stem cell (iPSC) derived osteoblasts in a 3D, plasma treated, porous, gelatin coated, poly-L-lactide scaffold, we generated a 3D bone tissue culture with extracellular calcium deposition. By applying cyclical compression with an Arduino actuation device, we observed an improved bone phenotype. This method of generating 3D bone tissue may serve as a fast and accessible model for the formation of both healthy and pathological bone.

David Mack Lab at the Institute for Stem Cell and Regenerative Medicine

Identifying subsets of phenotypes in iPSCs and their differentiated progeny allow us to optimize tissue models for research. Here, we generated a rainbow reporter line in iPSCs that can track individual cells as they clonally expand and differentiate while providing phenotypic information. Knocking in four copies of a cassette containing three distinct fluorescent proteins allowed the expression of up to eighteen different colors. To date, we were able to create eight color barcoded iPSC lines for further experimentation. Color coded cells will allow us to track how the initial location/physical stresses/phenotype of an iPSC-derived cardiac cell in an engineered tissue determines its tissue layer and cell type.

Cell Biomechanics Lab at the Institute for Stem Cell and Regenerative Medicine