Projects

  • Description: Developed a web application that allows users to upload text materials, create tests, conduct evaluations, and view feedback. Evaluations are conducted based on defined metrics. Users can also track their performance over time.
  • Role: Project Manager.
  • Technologies: LangChain, OpenAI API, Django, PostgreSQL, Plotly, Panel.
  • Duration: June 2024 - August 2024.
  • Team size: 8.
  • Description: Implemented 5 graph neural network models: GCNConv ,GRU-GCNConv, GCNConv-LSTM, MPNN, GAT-Transformer on time series data of water depth levels in a Storm Water Management Model (SWMM).
  • Role: Sole Developer.
  • Technologies: PyTorch Geometric, Matplotlib, Plotly.
  • Duration: January 2024 - April 2024.
  • Team size: 1.
  • Description: Developed a pipeline for identifying significantly differentially expressed proteins in proteomics datasets using a hybrid Frequentist-Bayesian approach. The workflow first applies Frequentist statistical tests (ANOVA, Welch’s test, Kruskal-Wallis) to screen significant proteins, followed by Bayesian inference for verification. It includes batch normalization, missing data handling, and log2 fold change analysis, ensuring robust results. The pipeline also integrates visualizations such as volcano plots and Bayesian posterior distributions to compare statistical methods and enhance interpretability.
  • Role: Sole Data Scientist.
  • Technologies:PyMC, Statsmodels, ArviZ, Scikit-learn, Pandas, Numpy, Matplotlib, Seaborn.
  • Duration:July 2021 - August 2021.
  • Team size:1.
  • Description: Implemented various deep learning models, including CNN, RNN, Transformer, BERT, and hybrid models, to identify the optimal model for predicting 16S rRNA gene copy numbers. The goal was to achieve higher accuracy than traditional bioinformatics and phylogenetic tools. Results indicated that the best model was BERT, with an RMSE of approximately 0.5, surpassing traditional tools with RMSEs ranging from 2-3.
  • Role: Graduate Research Assistant.
  • Technologies: TensorFlow, PyTorch, Scikit-learn, Matplotlib.
  • Duration: September 2023 - August 2024.
  • Team size: 2.
  • Description: The unbinding rate of Hoechst 33342 is not constant but depends on the cell growth rate. This dependence is mediated by cellular activity, forming a feedback loop with the inhibitor’s activity. There is cell-to-cell heterogeneity in inhibitor-target interaction, leading to the coexistence of two distinct subpopulations: actively growing cells that dissociate the inhibitors from the targets and non-growing cells that do not.
  • Role: First Author.
  • Methodology: Bacterial culture, bacterial strain construction with KEIO library, small-molecule binding/unbinding rate measurement, time-lapse imaging, β-Galactosidase assay, fluorescence microscopy, image analysis with Fiji, statistical analysis with R, SPSS.
  • Duration: February 2020 - January 2022.
  • Team size: 4.
  • Description: E. coli cells responded heterogeneously to protonophores, resulting in bimodal distributions of cell growth, substrate transport, and motility. This heterogeneous response required active efflux systems. The response is driven by efflux-mediated positive feedback between PMF and protonophores’ action.
  • Role: First Author.
  • Methodology: Bacterial culture, strain construction with KEIO library, fluorescence microscopy, motor speed measurement.
  • Duration: September 2018 - January 2021.
  • Team size: 5.
  • Role: Second Co-Author.
  • Methodology: Bacterial culture, strain construction with KEIO library, smFISH, β-Galactosidase assay, automated image analysis with MATLAB, fluorescent labeling of proteins using L-azidohomoalanine.
  • Duration: September 2017 - March 2018.
  • Team size: 4.
  • Description: Eradicating bacteria with antibiotics is stochastic in nature. Bactericidal antibiotics induced population fluctuations. At high concentrations, bacterial clearance dynamics were heterogeneous. At low concentrations, clearance still occurred with a non-zero probability. Eradication could be facilitated by increasing extinction probability.
  • Role: Third Co-Author.
  • Methodology: Bacterial culture, strain construction with KEIO library, MIC measurement by plate assay, time-lapse microscopy, replicate culture by microtiter plate, image analysis with MicrobeJ.
  • Duration: March 2017 - September 2017.
  • Team size: 6.
  • Description: Distributions of membrane proteins differ in embryonic, post-natal, and mature mouse RPE, suggesting developmental regulation of protein trafficking. Loss of Tsg101 severely disturbed the polarity of RPE, which forms irregular aggregates exhibiting non-polarized distribution of cell adhesion proteins and activation of epidermal growth factor receptor signaling.
  • Role: First author.
  • Methodology: mating of genetically-engineered mice, immunohistochemistry, electron microscopy, electroretinogram (ERG) and optomotry, statistical analysis.
  • Team size: 9.