SEE: A software
supported Strategy for the Exploration of Exploration
SEEpro: SEE in the Mathematica environment
|The path traced by a C57BL/6J mouse during a 30 min session in the Open Field (A) is partitioned by the SEE algorithms into four intrinsically distinct components (B) (from Lipkind et al., in preparation).|
A most intriguing question in the study of the genom/brain/behavior interface is that of the appropriate representation of behavioral phenotypes. The segmentation of behavior into behavior patterns (Fixed action patterns, behavioral categories) either manually, or by using machine learning-based systems that enable researchers to encode their intuitive manual labels into behavior detectors that can then be used to automatically classify the behavior in large data sets with high throughput is now widespread. This is a straightforward way for establishing comparing the effects of pharmacological treatments, finding differences between genetic preparations, epigenetic influences and gender, and for comparing frequencies of these categories in closely related species and strains. This way, is however not useful for comparisons across species and preparations that are not closely related. For such comparison it is necessary to use low level descriptions that are more likely to apply universally. Moreover such comparisons have to be generalizable, in the sense that discoveries arrived at one laboratory would be replicable in other laboratories.
Our response to this challenge is SEE, a software supported methodology
for the measurement of the kinematics (geometry) of whole animal free movement.
We have developed dynamic representations and measures (also termed in
behavior genetics behavioral endpoints) that are:
1) based on intrinsic geometric and statistical features of the behavior, and
2) are not released unless first tested for replicability across several laboratories.
The current form of SEE is a result of an ongoing effort to which many people
have contributed for the past 12 years. The first ethologically relevant pattern
on which SEE was later based - the home base phenomenon – was documented in 1975
by Paul Nau and Ilan Golani (Ph.D. thesis submitted by Nau to the dept. of
Psychology, Dalhousie University N.S. Canada). This and related phenomena (slow
outbound - fast inbound portions of excursions, stop-and-go behavior) were
quantitatively documented by David Eilam (1989,1993).
Ofer Tchernichovski first
introduced automated tracking (1995), used it for the computation of
moment-to-moment velocities and developed several measures that are now central
to SEE (1996,1998). The first software version of SEE in the MathematicaTM
programming environment was developed by Dan Drai, who also gave it its name
(2001). Drai also developed the algorithm segmenting the animal’s path into
segments of progression and stopping, on which many of the SEE endpoints are
based (2000). All these studies were performed in the dept. of Zoology of Tel
Aviv university, towards Ph.D. degrees, in collaboration with and under the
guidance of Ilan Golani (Zoology) and
Yoav Benjamini (Statistics).
Neri Kafkafi used SEE for analyzing rat behavior in a photobeam cage (2001) and wrote
Experiment Explorer and Endpoint Manager - two MathematicaTM software
extensions of SEE (2002).
Further developments of SEE, including the four C++ stand-alone programs available on this website,and the continuing development of new algorithms and behavioral measures (endpoints) have been and have been implemented by Liad Shekel (Statistics), and a version of SEE in R has been written by Itamar Eskin (RSEE), both with the support of an ERC grant (PSARPS).
In the past, our team included zoologists (Ilan Golani,
Dina Lipkind, Guy Horev, Anna Dvorkin, Ehud Fonio, Eran Polosetski, Eyal
Gruntman, Tel Aviv and Neri Kafkafi
), statisticians (Yoav Benjamini, Dan Yekutieli, and
Anat Sakov, Tel Aviv), and a neurobiologist (Greg Elmer, The Maryland Psychiatric Research Institute-MPRC).
See our old team picture During the years 2012-2014 SEE has been elaborated and implemented in R by Liad Shekel and Itamar Eskin. A new version of SEEPro has been written by Neri Kafkafi in 2014.
Because the mouse has become the experimental animal par excellence of Behavior Genetics, our work is now centered on mice. In work already done by us and by others, it has been shown that contrary to what was common belief, rodent free locomotor behavior is quite structured. In the last decade we have developed SEE, for analyzing this structure. This software can visualize and quantify the patterns, using as input the automatically digitized time-series of the animal's location and producing as output, key parameters that characterize the behavior.
The ethologically relevant parameters isolated and measured by SEE are relatively independent of each other, and reveal a natural structure that is at least partly independent of the animal's level of activity. They reflect processes involving locomotion, motivation, navigation, spatial memory and learning. They can be measured automatically and efficiently in very large numbers of mice (high throughput) and their listing yields an objective algorithmic definition of species- and strain-specific behavior.
One feature that characterizes our methodology is extensive treatment of the raw data before the calculation of these behavioral endpoints. This treatment includes a sifting out of noise (performed by SEE Path Smoother). This software component is recommended for use on any time series of location data, regardless of whether one is going to subsequently use the other components of SEE. Following smoothing comes the segmentation into progression segments and stopping episodes (performed by SEE Path Segmentor). Finally, SEE Files Creator and SEE Endpoint Calculator are used for calculating a long list of ethologically relevant behavioral endpoints.
The complexity of exploratory behavior creates a need for a visualization and analysis tool that will highlight regularities and help generating new hypotheses about the structure of this behavior. The hypotheses can then be formulated as algorithms that capture the patterns and quantify them. SEEpro is a Mathematica based software developed by us for the exploration of exploratory behavior. The raw data for SEEpro are a time series of the animal's coordinates in space sampled at a rate that allows a meaningful computation of speeds.
These visualizations highlight the presence of preferred places, including the animal's so-called home base, and permits a computation of the spatio-temporal diversity in the location of stopping episodes.
The software also:
Produces a visualization of the way places in the animal's operational world are connected to each other.
SEEpro also permits the definition and computation of behavioral endpoints across any section of any database of raw data. The range of applicability of SEEpro to various experimental setups, tracking procedures, species, and preparations is addressed in the discussion.