Predictive Platform for Fast Antibiotic Susceptibility Test

Introduction

I was Algorithm Engineer Intern for SeLux Diagnostics Inc, a biotechnology company, where the team are building the next generation high throughput FDA required antibiotic susceptibility testing (“AST”) device for clinical labs to run AST test for infectious disease efficiently and precisely.

Quite amount of data were generated from bio-experiments, chemical-experiments, device running, patient samples, I used OOP to capture those information from various objects such as physical device parameters, incubation results of microbiological plates, chemical reagents fluorescence value, bacterial genera, drug concentration, patients’ record.

Then I investigated their relationship, and developed and applied machine learning algorithms (Kernel Bayes, XGB, hmm, Bayes nets with Chow-Liu tree, conditional random fields, and deep learning LSTM) to do different predication, such as minimum antibiotic susceptible concentration for different sample.

I also participated in built a multi-functional software to embed the algorithms into the system of device for seamlessly backend device running and algorithm training, testing, and frontend web-visualization using the multi-threads pipe and ASP.NET web service to communicate data between device, database, and algorithm development platform (such as python, tensorflow, Matlab, R)(github).