298
P. Sakagiannis et al.
tual model can be extended to address the above claims by adding tonic and/or
phasic input.
4.2 Intermittency Can Extend Functional Models of Larva
Locomotion
Traditional random walk models fail to capture the temporal dynamics of animal exploration [5]. Even when time is taken into account in terms of movement speed, reorientations are assumed to occur acutely. Integrating intermittency can address this limitation allowing for more accurate functional models
of autonomous behaving agents. Such virtual agents can then be used in simulations of behavioral experiments promoting neuroscientifically informed hypothesis that advance over current knowledge and generate predictions that can stimulate further empirical work [1].
It is widely assumed that Drosophila larva exploration can be described as
a random walk of discrete non-overlapping runs and reorientations/head-casts
[9] or alternatively that it is generated by the concurrent combined activity of
a crawler and a turner module generating repetitive oscillatory forward peristaltic strides and lateral bending motions respectively and possibly involving
energy transfer between the two mechanical modes [3,6,13]. Both models can
easily be upgraded by adding crawling intermittency which might or might not
be independent of the lateral bending mechanism. In the discrete-mode case,
intermittency can simply control the duration and transitions between runs and
head-casts or introduce a third mode of immobile pauses resulting in a temporally
unfolding random walk. In the overlapping-mode case the two modules are complemented by a controlling intermittency module forming an interacting triplet.
Depending on the crawler-turner interaction and the effect of intermittency on
the turner module, multiple locomotory patterns emerge including straight runs,
curved runs, stationary head-casts and immobile pauses. This simple extension
would allow temporal fitting of generative models to experimental observations
in addition to the primarily pursued spatial-trajectory fitting, facilitating the
use of calibrated virtual larvae in simulations of behavioral experiments.
4.3 Limitations
A limitation of our study is that due to the single-spinepoint tracking, it is
impossible to determine whether micro-movements occur during the designated
inactivity periods, an issue also unclear for adult fruitflies in [12]. It follows
that in our analysed dataset and in [12], immobile pauses, feeding motions and
stationary head casts are indistinguishable. Therefore, what we define as rest
bouts should be considered as periods lacking at least peristaltic strides but not
any locomotory activity whatsoever. Our relatively low velocity threshold V θ =
0.085 mm/s though allows stricter detection of rest bouts as it is evident from
the higher activity ratio (higher than 0.7 in most control groups in comparison
to lower than 0.25 in [12]). To tackle this, trackings of higher spatial resolution
with more spinepoints tracked per larva are needed, despite the computational
challenge of the essentially long recording duration.
P. Sakagiannis et al.
tual model can be extended to address the above claims by adding tonic and/or
phasic input.
4.2 Intermittency Can Extend Functional Models of Larva
Locomotion
Traditional random walk models fail to capture the temporal dynamics of animal exploration [5]. Even when time is taken into account in terms of movement speed, reorientations are assumed to occur acutely. Integrating intermittency can address this limitation allowing for more accurate functional models
of autonomous behaving agents. Such virtual agents can then be used in simulations of behavioral experiments promoting neuroscientifically informed hypothesis that advance over current knowledge and generate predictions that can stimulate further empirical work [1].
It is widely assumed that Drosophila larva exploration can be described as
a random walk of discrete non-overlapping runs and reorientations/head-casts
[9] or alternatively that it is generated by the concurrent combined activity of
a crawler and a turner module generating repetitive oscillatory forward peristaltic strides and lateral bending motions respectively and possibly involving
energy transfer between the two mechanical modes [3,6,13]. Both models can
easily be upgraded by adding crawling intermittency which might or might not
be independent of the lateral bending mechanism. In the discrete-mode case,
intermittency can simply control the duration and transitions between runs and
head-casts or introduce a third mode of immobile pauses resulting in a temporally
unfolding random walk. In the overlapping-mode case the two modules are complemented by a controlling intermittency module forming an interacting triplet.
Depending on the crawler-turner interaction and the effect of intermittency on
the turner module, multiple locomotory patterns emerge including straight runs,
curved runs, stationary head-casts and immobile pauses. This simple extension
would allow temporal fitting of generative models to experimental observations
in addition to the primarily pursued spatial-trajectory fitting, facilitating the
use of calibrated virtual larvae in simulations of behavioral experiments.
4.3 Limitations
A limitation of our study is that due to the single-spinepoint tracking, it is
impossible to determine whether micro-movements occur during the designated
inactivity periods, an issue also unclear for adult fruitflies in [12]. It follows
that in our analysed dataset and in [12], immobile pauses, feeding motions and
stationary head casts are indistinguishable. Therefore, what we define as rest
bouts should be considered as periods lacking at least peristaltic strides but not
any locomotory activity whatsoever. Our relatively low velocity threshold V θ =
0.085 mm/s though allows stricter detection of rest bouts as it is evident from
the higher activity ratio (higher than 0.7 in most control groups in comparison
to lower than 0.25 in [12]). To tackle this, trackings of higher spatial resolution
with more spinepoints tracked per larva are needed, despite the computational
challenge of the essentially long recording duration.