Transcript alix_-_mcn2011_presentationx

Predicting responses to stimuli
in larval zebrafish
Alix Lacoste
Methods in Computational Neuroscience
Final Project Presentation, August 2011
Arousal in larval zebrafish is correlated with rest
Stimulus Stimulus
Prober et al. 2006
Fish are more active during the day
than at night
Probability
of
Response
during the
day
at night
Increasing stimulus (tap) strength
The response to their environment varies
with stimulus strength and circadian time
Analyzing responses to stimuli
1. Predicting response to stimuli using the history of movement
• Can movement before the stimulus (tap) predict whether the fish will
respond to the tap?
• Hypothesis: If fish have not moved recently, they are less likely to
respond to the tap
2. Classifying fish by response types
• Are there different types of responses to the stimulus?
• Hypothesis:
1. Movement before a tap
influences response probability
Half of the tap trial data = training set
Mean Amplitude of movement
(Δ pixel)
Average trace before a tap, conditional
on the response or lack of response
Other half of the tap trial data = test set
Projection
onto trace
followed by
response
Projection
onto trace
not followed
by response
Time before the tap (sec)
Movement prior to a tap can predict
whether the fish responds: example
y
y
x
x
y=
Projection of test
movement vector onto
the mean training
vector that is followed
by no response
Projection of test movement vector
x = onto the mean training vector that is
followed by a response
Movement prior to a tap can predict
whether the fish responds: more examples
• Responses and non responses are correctly separated with this
analysis
• In cases where there was no movement before the tap, it is not
possible to predict whether the fish will move in response to the tap
2. Response clustering
Eigenvalues
Movement
amplitude
PCA
Main Eigenvectors
Time relative to the stimulus (tap)
Response clustering
Projection of post-tap movements
onto main eigenvectors
Average response waveforms
Group 1
Group 2
• Principal component analysis finds the very large responses (group 1).
These responses are likely initiated via a distinct neural circuit
(Mauthner cell escape system?)
Conclusions
1. We can predict whether a fish will respond to a stimulus
by analyzing its movement prior to the stimulus
• Next steps: Use more sophisticated statistics of prior
movement to better predict responses
2. Response movement traces can be clustered using
principal component analysis and suggest that distinct
neural circuits are involved in producing responses
• Next steps: a) Use movement statistics to predict
response type. b) Find neural correlates of different
response types.
Thank you!
David Schoppik in Alex Schier’s lab
Hubert and Elad
Michael Berry