Experimental Design

We analyzed and utilized single cell flow cytometry data with the final goal of inputting it into machine learning models in order to diagnose the presence of leukemic cells. In order to identify differentially expressed heterogeneous cancer clusters in each patient we will be using FlowSOM to identify the divergence of expression of meta clusters from healthy and cancer patients. We then normalize our data if needed, using distinct clustering techniques (UMAPs and K-means) to separate different cell types, and visually representing our data through histograms. We designed a supervised model (Random Forest) and an unsupervised model (Deep Learning). The training for both types will be different as the supervised models will use and process the raw Flow Cytometry data, meanwhile the unsupervised models will take the input of UMAP images for classification. The binary classification models will return 0 for healthy patients and 1 for patients with leukemic clusters.