Jiawei Zhao

This is probably one of the greatest apps ever created.

I started this project as a way for learning React and it has since grown into a fully fledged & fullstack academic web-app. I developed this web-app for protein structure prediction and 3D molecular visualization using JSmol independently at Vihinen Lab of medical faculty, Lund University.

The web-app was developed with React, Typescript, Redux and GraphQL.

A detailed decription is at https://github.com/Orthologues/BINP39-Visualization-Project/tree/production/react_ts_express_app.

Aside from Front-end development, I also developed server-side scripts independently in Node.js, Express.js and Typescript to run python sub-process for protein structure prediction, PDB file generating and residue mapping. File transfer, Nodemailer and RESTful routes were implemented in my backend scripts. The web-app is deployed on a CentOS-7 server. I used NGINX reverse proxy for deployment as well as pm2 for process management.

Let’s look at some exciting screenshots of VIEW-SC+

Pipeline of residue-level PDB to UniProt variant mapping

This was part of my BINP29 course in master's program in Bioinformatics at Lund University.

A local app run on command line, which uses either Needleman-Wunsch algorithm or Smith-Waterman algorithm to align nucleotide or amino acid sequences. Run SoftwareUI.py to start the appication. ‘tkinter’ has to be installed as one of your python3 packages before running the script.

This was my bachelor thesis project at Huazhong University of Sci & Tech.

Standardized and processed raw ultrasound images of carotid arteries with artificially predefined labels usingMatlab in order to construct the Data set. Used Tensorflow to build a CNN of Supervised Learning based on a paper at https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5293622/, trained the Data set using 5-fold cross validation, adjusted its parameters and the network structure for automatic classification of the ultrasound images into threes types: plaque, thickened and normal. The average precision value of every category (plaque, thickened, normal) of my validation set acquired by my CNN is higher than its corresponding average precision value from the method of Support Vector Machine (SVM) by more than 5%.