Machine learning discriminates parkinsonian movement disorders in zebrafish

Parkinson’s disease is caused by a progressive loss of dopamine neurons in
the substantia nigra. A loss of motor control occurs in patients, with symptoms
including bradykinesia, resting tremor and muscle rigidity. There is currently no
cure for Parkinson’s disease and the commonly used treatment, L-DOPA, can
cause side effects including dyskinesia. To identify potential treatments, drug
screening using a suitable animal model is required before reaching clinical
trials. Zebrafish are a vertebrate model organism well suited for high throughput
drug screens, and genome editing can be used to create heritable mutations in
causative genes to model human disease.
This thesis presents five genetic models of Parkinson’s disease in the zebrafish
created by CRISPR/Cas9 targeting (pink1, parkin, dj-1, fbxo7, and gba).
Molecular analyses show a loss of dopamine neurons in the brain of the DJ-1
deficient zebrafish. This thesis also presents findings from a transcriptomic
analysis of the dj-1 mutant brain revealing dysregulated genes consistent with
known parkinsonian defects.
An important focus of this work is the development of a novel computational
method to analyse the movement phenotype of a zebrafish. The method
developed uses high-speed recordings of zebrafish swimming, processed with a
new fish tracking software. By measuring spatial coordinates and angles along
the spine, swimming movement was converted into data suitable for
computational analysis. The movement data was subjected to unbiased
analysis employing a white-box supervised machine learning method (an
Evolutionary Algorithm) which successfully discriminated the dj-1 mutant from
wild type.
This thesis concludes that the DJ-1 deficient zebrafish is a representative
animal model of Parkinson’s disease, and that machine learning can be used to
classify the model based on movement data alone. It is proposed that the
techniques developed here, have the potential for drug screens on the dj-1
mutant using evolved classifiers to assess treatment effectiveness.