Work

Bursti-tunnistusmenetelmät ihmisperäisistä monikykyisistä kantasoluista erilaistetuille hermosoluverkostoille

Year: 2019
Type: dissertation
Abstract: A burst is a set of subsequent action potentials that are fired at a high frequency. Although bursts are a fundamental part of electrical activity of neuronal networks in vitro, no standardized method exists for burst detection. Visual identification of bursts is a widely accepted method, but it is not objective nor time-efficient. Therefore, various algorithms have been developed for burst detection. Burst detection algorithms are typically developed and verified only on one specific type of data. This can be problematic because the bursting activity is highly variable between different cell types. Consequently, the applicability of the algorithms is restricted to a narrow range of activity types. Especially applicability to human neuronal networks is questionable because the algorithms are often developed on rodent neuronal networks, which display distinct activity patterns in comparison to human networks. The aim of this thesis was to produce a test data set, which would well represent bursting and non-bursting activity observed in human pluripotent stem cell (hPSC)-derived neuronal networks, and to identify a single algorithm with optimal parameters that would successfully detect the bursts in this test data set. As rodent neuronal networks are also widely used in neuroscience, the algorithm was desired to function also on activity derived from rodent cultures. To achieve these goals, hESCs were differentiated into functional neuronal networks and cultured on microelectrode array (MEA). Primary rat cortical neurons were similarly cultured on MEA. Electrical activity of the developing networks was recorded twice a week until synchronized bursting emerged. At this point, pharmacological assays were performed in order to record modulated activity. On the MEA recordings, distinct activity patterns were identified, and short recordings representative of the distinct patterns were included to the test data set. The performance of four contemporary burst detection algorithms was evaluated on the test data set. The evaluation was based on visual identification of bursts from the raw MEA signal. For each algorithm, a performance score was determined and sensitivity and specificity were computed. The evaluation was performed in two runs using either 3 or 5 as minimum number of spikes required for a burst. Other algorithm parameters were set to default values suggested by the original authors. The optimization possibilities were not encouraging for other algorithms but logISI, which also provided the highest performance and the most balanced sensitivity and specificity values. Parameters of logISI were optimized for the test data set, which significantly improved its performance. As a result, logISI displayed good or excellent performance on the test data obtained from human and rat neuronal networks during spontaneous and pharmacologically modulated activity. Based on these results, logISI could have the potential to become a standard burst detection algorithm in the field.
Institution: Aalto University
Language: English
Open Access status: green
OpenAlex ID: W7148204504
Cites: 59
Cited by:
Related to: 10