Whole blood transcriptomics has led to a deeper understanding of the underlying biology of
human disease. These studies traditionally compare gene expression at the time of sampling
between subjects with different disease causes and manifestations. More recent
bioinformatic approaches have enabled the use of a previously untapped component of the
transcriptome, unspliced transcripts, to study the dynamics of many biological processes.
RNA velocity analysis, originally developed for scRNA-Seq, has expanded the use of
unspliced transcripts, along with spliced transcripts, to calculate the direction and
magnitude of gene expression change and model future transcriptomic trajectories. Given
that a patient’s “current” whole blood transcriptome is representative of their current
disease state, I hypothesised that future transcriptomic states (as modelled using RNA
velocity) will be representative of future disease states. Thus, the goal of this project was to
adapt RNA velocity analysis to predict future disease states of individual subjects. To achieve
this, I developed a pre-processing pipeline that quantifies spliced and unspliced transcript
expression from raw reads in whole blood RNA-Seq datasets. The RNA velocity analysis
algorithm was also adapted to analyse these datasets and further developed to assign
probabilities of transition of subjects to different biological states. These methods were
firstly applied to reconstruct the expression dynamics of two relatively simple biological
systems in time-course datasets, revealing the importance of certain factors on predictions,
including gene and hyperparameter selection. I then applied this analysis to a large cohort
of severely sick African children to predict disease outcomes (survival or death). RNA
velocity analysis performed similarly to more conventional prediction methods. Therefore, I
have illustrated proof of principle of a novel approach that predicts disease outcomes from
whole blood transcriptomics.