BSI+Google

class: left,

name:opening

## Data Intensive Brain Science

Joshua T. Vogelstein
<br>
Kavli Neuroscience Discovery Institute

<br>
questions: [jovo@jhu.edu](mailto:jovo at jhu dot edu)
<br>
slides: <http://brainx.io/BSI-google>
<br>
funding: DARPA I2O {GRAPHS, XDATA, D3M, L2M}

.center[
<img src="images/biomedical-engineering.small.horizontal.white.png" STYLE="HEIGHT:100px;"/>
]

---
class: center, middle

.h1[# <center>Motivation</center>]

---
class: center, middle

<footer>source: NAMI</footer>

---
class: left

## The Human Condition

---
class: left

## The Grazing Goat Starves

---
class: left

## Our Social Cage

---

# Goal

Give each individual the tools she needs to move herself in the  desired direction by the desired amount in our high-dimensional experience

---

class: center, middle

.h1[# <center>Challenges</center>]

---
class: center, middle

---
class: center, middle

---
class: center, middle

---
class: left, middle

# Enter NeuroData

---
class: left

## Our Idea

Multiple measurements from same subject should be more similar to one another than they are to any measurement of any other subject

$P[|| x\_{ij} - x\_{ij'}|| < || x\_{ij} - x\_{i'j''}||]$

<br>

Without that, how can we trust  biomarkers?

(we prove a bunch of stuff about our estimator)

---
class: top, left

## Our model

---
class: top, left

## Our Pipeline

---
class: center, middle

---
class: top, left

## Our Result

Within a study, subjects are "discriminable"

.pull-left[
- 25+ studies
- 6000+ scans
- largest "meganalysis"
- largest open repo
]

.pull-right[
<img src="images/discr.jpeg" style="height: 100%;"/>
]

(we tried 200+ pipelines ⇒  1M+ compute hours)

---
class: top, left

### But...

-  Across studies, populations are significantly different
- Conditioning on Phenotype Fails

---
class: top, left

### Our Proposed Solutions

- better pipelines
- better data (eg, quantitative MRI)
- deep phenotyping
- data acquisition harmonization

---
class: top, left

## Acknowledgements

---

### Questions?

- dimensionality  reduction: [LOL](https://github.com/neurodata/LOL)
- classification &  regression: [RerF](https://github.com/neurodata/R-RerF/)
- hypothesis testing: [MGC](https://github.com/neurodata/mgc)
- clustering: [knor](https://github.com/neurodata/knorr)
- email: [jovo@jhu.edu](mailto:jovo@jhu.edu)
- lab:  [neurodata](http://neurodata.io/)
- startup: [gigantum](http://gigantum.com/)

<br>