document.querySelector('video').playbackRate = 1.2
Dr. Taylor’s Tao of data analysis:
Follow the data, and abstract as little as
possible!
Occasionally, thoughtful abstraction and summary statistics will be needed and helpful, but much more rarely, and usually only in end-stage analysis or automation, not in initial exploration (initial bushwhacking science).
Early studies
* Lesions (accidental and otherwise)
* Neurosurgery - lesion and direct stimulation
* PET studies
* fMEG, fMRI
Do regions perform modular functions?
What is “representation”?
Function/Representation in cortex (task fMRI)
Classic fMRI
Define: Activation
Estimating the reproducibility of psychological science = 35%
(OpenScience-Collaboration, 2015; Science)
Single study fMRI: What stinks?
Does anyone know what kind of animal this is with “significant”
activations in its brain?
“There is increasing concern that most current published research
findings are false. In this framework, a research finding is less likely
to be true when the studies conducted in a field are smaller; when
effect sizes are smaller; when there is a greater number and lesser
pre-selection of tested relationships; where there is greater
flexibility in designs, definitions, outcomes, and analytical modes;
when there is greater financial and other interest and prejudice; and
when more teams are involved in a scientific field in chase of
statistical significance. Simulations show that for most study designs
and settings, it is more likely for a research claim to be false than
true. Moreover, for many current scientific fields, claimed research
findings may often be simply accurate measures of the prevailing bias.”
e.g. irradiating infant thymuses, statins, social psychology, nutrition
research, Alzheimer’s drugs, etc.
Academics (people) are not so rational about distributing research efforts or money to problems.
Red = popularly studied (Behrens, 2012)
Impact factor correlations (Behrens, 2012)
Picking the wrong brain region is a bad career move…
Manual meta-analysis of function performs ok…
What enabled the industrial revolution?
Craftsperson to Assembly line.
=> 
Large scale cooperation requires standardization.
(GDP)
Can standardization speed scientific progress as well?
Craftsperson scientist: Single study versus ?
=>
=> ??
Solution 2: Very large projects
WU-Minn-Oxford group (the good one of the pair of schools doing this
project)
* Structural data were used for connectivity (above)
* Functional data used for meta-optimization (upcoming)
Blue Brain Project Digital reconstruction of the brain by
reverse-engineering mammalian brain circuitry
Blue Brain Project
Maps the expression of EVERY gene in the entire brain
Requires data sharing, centralized repositories
Formal ontology for neuroscience studies
Computability!
BrainMap: 20 years of formally coded fMRI studies in one database
Reminder: activations
Representation in human cortex BrainMap hierarchical clustering of behavioral labels by activation locations alone:
Functional networks in human cortex Functional activations (1000s of studies) versus functional connectivity (1 study)
Side note: diseases show increased prevalence at cortical network hubs, including Alzheimers dementia, Aspergers syndrome, schizophrenia, frontotemporal dementia, juvenile myoclonic epilepsy, progressive supranuclear palsy, left and right temporal lobe epilepsy, and post-traumatic stress disorder.
Neurosynth: automated fMRI databasing Neurosynth platform (backend in Python) auto-extracts tabular fMRI activation coordinates and word frequencies from published studies:
Representation in human cortex
* Forward inference maps show the probability of activation given the
presence of the term, P(act.|term)
* Reverse inference maps show the probability of the term given observed
activation, P(term|act.)
Data analysis task:
Which word-activation associations in the neurosynth database best
spatially match your current brain state?
a.k.a. Mind-reading
Neurosynth: Mind reading task
Classifiers:
General method we will cover in class – many very cool types!
This one is just a naive Bayes model.
Above chance (diagonals) for every category (i.e., success!)
Validated by manual meta-analysis
How has the function of regions been studied?
Early studies
* Lesions (accidental and otherwise)
* Neurosurgery - lesion and direct stimulation
* PET studies
* fMEG, fMRI
Modern neuroinformatics and computational
neuroscience
* Large databases of studies:
* fMRI, MEG, DTI, rfMRI, gene expression, neuronal stucture, cellular
connectivity
* Increases in power (n)
* More robust to bias (not entirely)
* Computable ontologies
* Functional models as hypotheses and publications/literature
Goal is to make this model-building process much more systematic:
Not 1 hypothesis, but at least 2, or better yet, systematically refining a computational model (e.g., bi-weekly model refinement based on empirical data).
The model is the knowledgebase, and should be the unit of publication, at least in many domains.