Home  
  Spike Train Toolkit
     - Primer
     - Introduction
     - Toolkit Wiki
     - Download
     - Demonstrations
     - Input Format
     - Input Structures
     - Output Structures
     - Information Options
     - Entropy Options
     - Function Reference
     - Version History
     - License
     - Contribute
  Analysis Overview
     - Analyze Data
     - Your Submissions
  Neurodatabase
  BrainML Models
  LNI Homepage
 
User ID
Password


  What will the Spike Train Analysis Toolkit tell me about my data?



A Primer on Information Theory and the STAToolkit

What can STAToolkit tell me about neural coding?
Neural coding—the representation and processing of information with spike trains—is fundamental to sensation, perception, decision, and action, and can become dysfunctional in disease states. However, the features of neuronal activity that convey and manipulate information are not yet known. The downloadable Spike Train Analysis Toolkit (STAToolkit) provides a suite of analytic algorithms that enable asking specific questions about how real spike train data—recorded from any of a large range of preparations and protocols—represents or codes specific features of a stimulus or motor action. STAToolkit is a tested, adoptable, documented suite freely available for open source download and use by typical neurophysiology labs. Guides and tutorials, such as these pages, minimize the effort needed to apply these analyses appropriately.

What can STAToolkit tell me about my data?
STAToolkit routines enable exploring data by asking questions such as:
- What aspects of neural responses encode the stimuli that elicit them, or the behavioral context of: typical single unit data? long streams of continuous data? episodic data? bursty data?
- How do recorded spike trains resemble one another so as to form natural clusters in a response space?
- What neural codes are plausibly—or implausibly—represented by recorded activity?
Planned extensions to STAToolkit (see below) greatly expand this list.

How do I start using STAtoolkit?
See the Getting started section on the Spike Train Analysis Toolkit Introduction page.

Some of my spike data are multineuronal. What can STAToolkit do for my data?
Several of the STAToolkit methods are suitable for multineuronal data. See the documentation.

I have LFP or evoked potential data. What can STAToolkit do for my data?
Our input data format and input data structures can handle both episodic and continuous data. Currently, only the binless methods can handle continuous data analysis. Modifications of other STAToolkit methods for continuous data are planned for the near future (see table below).

I'm a neurophysiologist, not a computational neuroscientist. Can I use STAToolkit and how much help will I need?
Because we designed STAToolkit for the neurophysiologist with a need to perform new analyses that pose or test hypotheses relating to data, the package includes clear and readable user documentation for verifying hardware and software compatibility, installation, use, and applicability. In designing and producing this documentation, we have tried to maintain a neurobiologist's perspective, consistent with our goal to enable neurophysiology labs to select as well as carry out information-theoretic explorations of their data. The documentation also includes extensive literature references citing the originators of the various methods we have implemented in STAToolkit, both to guide proper use of the routines and also to properly acknowledge the intellectual property of each of those who devised such methods.

Why was STAToolkit designed?
STAToolkit was designed as a suite of tools for enhanced examination of neural coding, and it was designed to be adopted and used by investigators working in a wide range of preparations. The motivation was to facilitate greater use of information-theoretic methods that can relate neural firing to sensation or behavior. Towards data-driven theories of neural coding, STAToolkit routines allow testing of hypotheses such as:
- different regions, networks, or modalities within the nervous system utilize different neural codes,
- individual regions, networks, or modalities utilize different coding at different times, or in different contexts,
- mechanisms of neural or mental disorders may be better understood by their effects on neural coding.

How is STAToolkit used? By whom?
STAToolkit is in active use by the neuroscience community, with more than 1,200 downloads. We have an active user base, and STAToolkit methods have been applied to many preparations and protocols, including retinal coding, prehension in awake behaving primates, mammalian taste discrimination, thalamocortical spike timing, spike timing in neural networks, EOD analysis in electric fish, and various population codes. STAToolkit has been used in courses in both computational neuroscience and neuroinformatics. STAToolkit is being adopted by leading neuroinformatic resources in the US, the UK, and Canada, and our users have made and offered extensions and enhancements.

What is information? What is entropy?
Entropy measures the dispersion of the distribution of a single variable, usually calculated from a histogram of its values. Although we measure entropy towards calculating information, entropy is a quantity of interest in and of itself and the STAToolkit can provide it as an intermediate result. Information relates two variables (i.e., Shannon's mutual information); one is a measure of neural activity; the other is typically related to stimuli or behavior. Although information is defined and usually calculated as a difference in entropies, some of our methods bypass this step. We also distinguish between formal information and category-specific information. Formal information concerns all aspects of a response that depend on the stimulus. It is estimated from the difference between the entropy of responses to an ensemble of temporally rich stimuli and the entropy of responses to an ensemble of repeated stimuli. Category-specific information refers to the amount of information that responses convey about an experimental parameter that varies categorically (e.g., across several values of a stimulus or behavioral attribute).

What do you mean by categories? How do these relate to neural coding?
Categorization of stimuli or motor patterns allows relating neural firing and thus codes to an organism's sensation or action. Information theoretic analysis is central for exploring this, and STAToolkit allows segregating recorded datasets into any of several categories. Categorization and categories are fundamentally abstract and general, and can include relationships in time, different cortical areas, experimental vs. control, awake vs. anesthetized, wild type vs. genetically altered, as well as attributes of stimuli, behavioral sets, or motor responses associated with a recording. Note that information theory and the STAToolkit can also be used to rule out neural codes—by showing that the amount of information in measured neural activity is not enough to account for the sensory or behavioral choices.

Why is STAToolkit a suite of methods?
Elucidating principles of neural coding requires multiple analytic methods: specific neural systems are likely to use different kinds of representations, specific hypotheses require different measures as tests, and specific types and amounts of data require different algorithms for accurate analysis. STAToolkit therefore implements, documents, and guides application of several information-theoretic algorithms. These have distinct but overlapping domains of applicability, and they are complementary in their analytic strategies and in their assumptions about plausible neural codes. The straightforward way to distinguish biological differences from differences in methodology and an investigator's assumptions is to examine experimental neural data with a broad set of approaches, surmounting the pitfalls and confounds of relying on a single strategy by instead making use of a suite of algorithms that are complementary in their assumptions. STAToolkit is designed so that neurophysiologists can select the methods most appropriate for their experimental design, type and quantity of data, and hypotheses. Our guidelines free them from the need to precede analyses with time-consuming and detailed review of these methods.

Isn't one method best?
We do not anticipate that one or another of the algorithms we will implement will emerge as the "best", nor are they equally applicable. Rather, we offer them as complementary, spanning a gamut from the highly rigorous that require vast amounts of data, to those that require a range of a priori assumptions (e.g., about the topology of the response space, or the broad nature of the neural code) but less data. Only when multiple approaches yield corroborating results can one be sure that the conclusions reached are robust, not method-dependent, and the significance of the data maximized. Such a multifaceted analysis, not to mention selection among methods, is currently impractical for most investigators, but is uniquely enabled and publicly implemented by our project and its user tools.

Where can I download the toolkit?
You can get it here! For a complete list of system requirements, and installation instructions, please read this page from the STAToolkit Documentation.

Are there any examples or demos to help me get started?
Yes. Please read more about the included demos online, or in the included documentation.

Will the toolkit work on my system/platform?
Probably. Please see the list of platforms on which the toolkit in known to work. If your system is not listed, but you've got the toolkit running, please add it to the list. If you can't get the toolkit working on your system, we'd be happy to help.

What's ahead for STAToolkit?
Future extensions to STAToolkit, originated and supplied by our valued consultant-collaborators and now under development, will expand this analytic suite to provide:
- More information-theoretic methods targeting both single-unit neuronal activity and interactions of single neuron activity and population firing patterns, as well as calculation of additional information-like quantities useful for relating neural data to behavior.
- Beyond information theory, powerful techniques that analyze multineuronal firing patterns of an entire population, thus transcending standard approaches (such as cross-correlation) that analyze activity of two neurons at a time. Related techniques for dimensional reduction can identify relationships among multiple neurons, and across time, enabling more concise and incisive views of complex datasets.
- Several new computational approaches to assess synchrony and variability in neural populations. Synchrony of activity can report how a single underlying stimulus, transformed in different ways and recorded at separate locations, is bound to form a percept or plan an action. Analysis of responses to repeated stimuli can distinguish information-bearing patterns of variability from random ones, and suggest the biophysical origin of some fluctuations.
These are summarized in the Table below. In addition, the project plans:
- Procedures for generation of surrogate data sets can test specific hypotheses for population activity.
- More web-based tutorials and hands-on workshops will increase awareness, applicability, significance, and impact of STAToolkit.

Algorithm or method

Algorithm designer or consultant-collaborator

Spike train suitability

Multineuron suitability

LFP suitability

Status

Direct (Formal)

S.P.Strong, R. Koberle, R.R. de Ruyter van Steveninck, W. Bialek

***

**

 

in STAToolkit

Direct (Categorical)

S.P.Strong, R. Koberle, R.R. de Ruyter van Steveninck, W. Bialek

***

**

 

in STAToolkit

Metric Space

J. D. Victor, K. Purpura, D. Aronov

***

***

 

in STAToolkit

Binless Embedding

J.D. Victor

***

 

***

in STAToolkit

Context Tree

J. Shlens

***

Scheduled**

Scheduled*

in STAToolkit

Metric-Binless Hybrid

I. Nelken

***

**

*

Scheduled

Phase-Related Information

F. Sommer

***

*

***

Scheduled

Anthropic Correction

F. Theunissen

***

***

**

Scheduled

Generalized Information

S. Nirenberg, J. D. Victor

***

***

**

Scheduled

Multineuronal Dimensional Reduction

J. Shlens

***

***

 

Scheduled

Spatiotemporal Dimensional Reduction

C. Houghton, M. Magnasco, J.H.G. Dauwels

***

***

 

Scheduled

Surrogate Dataset Generation

M. Bethge, R. Brette, S. Shoham, M. Krumin

***

***

**

Scheduled

Stochastic Event Synchrony

J.H.G. Dauwels

***

***

***

Scheduled

Event Synchronization

R. Quian-Quiroga, T. Kreuz

***

***

*

Scheduled

Spike-Generation Variability

B. Knight

***

 

 

Scheduled

The table lists STAToolkit methods currently implemented and scheduled for development, including consultant-collaborators and/or originators of algorithms or code. All methods are suitable for spike train data; LFP or multineuron suitability is indicated by: ***-especially suitable; **-suitable; *-suitable with added preprocessing.

How is STAToolkit supported?
STAToolkit is entirely supported by Human Brain Project/Neuroinformatics grant MH068012, from NIMH, NINDS, NIA, NIBIB, and NSF. There is never charge to any of our many users. The project leverages methods developed by many consultant-collaborators under separate funding.




 Webmaster

Weill Medical College of Cornell University