Fig. 5: Predictive performance of machine learning models for disease severity.
From: Machine learning identifies T cell receptor repertoire signatures associated with COVID-19 severity
a AUROC curves for five machine learning models (gradient boosting trees, support vector machines, random forests, Bernoulli Naïve Bayes, and k-nearest neighbors) using 6-mer (left) and 3-mer (right) representations of TCR repertoire data. Models were trained to predict disease severity (moderate, severe) vs healthy donors for CD8 samples. Training and evaluation were performed using 500 iterations per model, average performance +/− 1 standard deviation shown on individual plots.
