Abstract
Impairments following traumatic brain injury (TBI) can be debilitating yet researchers have struggled to identify effective therapies. This is due to the incredible amount of heterogeneity presented by the patient population. Capturing this heterogeneity in a way that will advance our understanding of TBI will require data analytics able to harness and learn from the individual variability. In these studies, we assessed the feasibility of using both unsupervised and supervised machine learning (ML) techniques to uncover novel insights from the rat dataset collected by Operation Brain Trauma Therapy (OBTT). Initially, we utilized the combination of t-SNE (t-Distributed Stochastic Neighbor Embedding) and k-means clustering as an unsupervised approach to identify treatment groups. We observed treatment effects following intervention with a therapy that were missed in the univariate analysis. We then partitioned the dataset into acute input metrics (i.e., 7 days postinjury) and a defined end-study recovery outcome (i.e., memory retention) for the use of supervised ML. In numerous instances, the classifiers outperformed the baseline in their ability to predict end study cognition. Noting that treatment was impactful in determining model performance in this task, the next series of experiments assessed pairwise classifiers for their ability to predict drug administered when provided input from the full study. This uncovered that ML classifiers can accurately distinguish which pharmacotherapy an animal received. Lastly, we adapted the same ML workflow to analyze a synonymous dataset that differed in injury model. In this dataset, the classifiers struggled to distinguish between treatment in all instances except one. These results confirm that pharmacotherapies lead to unique recovery profiles following TBI and that these patterns differ depending on the injury model. This series of experiments both establishes the feasibility of incorporating ML into the analysis of laboratory TBI data and serves as a step towards identifying optimal treatments for specific TBI subgroups.