Software Development
Experience

MolecularNodes, a Blender plugin for molecular dynamics (MD) simulations, offers advanced rendering capabilities using industry-standard techniques. However, it lacks native solvent interaction visualization, crucial for understanding molecular structure-function relationships. This project integrates SolvationAnalysis, a plugin providing data structures for solute-solvent interactions, into MolecularNodes. The integration allows users to easily specify and visualize solvent interactions, appealing to researchers, students, and non-specialists without requiring programming expertise. The goal is to enhance MolecularNodes' functionality, providing professional-quality visualizations and promoting broader accessibility to advanced MD visualization tools.

• Code

Tinker9, an upgraded version of Tinker, will incorporate replica-exchange molecular dynamics (REMD) using C++ implementation and GPU acceleration. By running multiple replicas at different temperatures and exchanging configurations periodically, REMD enables more efficient exploration of complex molecular systems. This integration aims to overcome metastability and temperature-dependent stable states, streamlining molecular dynamics simulations and providing valuable insights into molecular behavior.

• Code

This project explores the use of an orthogonal scheme to solve the H-matrix and calculate total energy in Extended Huckel Theory for hydrocarbon systems. Challenges with carbon orbitals and atomic bonds were encountered, but promising results were obtained for diatomic hydrogen gas. Future improvements aim to refine the method's accuracy by revising the data structure, including internuclear distances, and enhancing the filtering mechanism.

• Code

We aim to showcase the power of deep learning in the realm of quantum mechanical (QM) Density Functional Theory (DFT) calculations. We focus on replicating the ANI-1 study, using a deep neural network (NN) to train on QM-DFT data. The main goal is to demonstrate how this approach can learn an accurate and transferable potential for organic molecules. By comparing the results with the original ANI-1 study, we seek to validate the efficacy and potential applications of the ANI-1 neural network model.

• Code

This project implements and evaluates a parallel accumulation algorithm for matrix multiplication to assess the advantages of parallelization. By comparing the performance of the parallel algorithm with the serial version and analyzing scalability with more threads, we aim to identify the most efficient matrix multiplication implementation. Performance analysis using "perf" focuses on cache miss ratios and instruction cycles to highlight the benefits of parallelization.

• Code

This project aims to develop accurate models for predicting the Air Quality Index (AQI) category and value at specific locations using weather and chemical data. We utilize logistic regression for AQI category prediction and linear regression for AQI value prediction. Model selection involves evaluating performance metrics such as mean squared error and accuracy. The project seeks to gain valuable insights into air quality patterns and support environmental monitoring efforts.

• Code