Princeton 2018

class: left,

name:opening

### A Community-Developed Open-Source Computational Ecosystem for Big Neuro Data

Joshua T. Vogelstein

.foot[[jovo@jhu.edu](mailto:jovo at jhu dot edu)
  |  <http://neurodata.io/talks/princeton-2018.html>]

---
class: center

### <center>What is NeuroData?</center>

### <center>Goal: Build tools that work on .r[your] data</center>

---

### Why is it hard?

- raw data are large collections of noisy 2D images
- too big to load into ImageJ,  MATLAB, or even Python
- maintaining local cluster is a pain
- sample sizes are small (e.g., 1)
- real data are nonlinear

---

### What is our solution?

- stop trying to build all the software ourselves
  - glue modular things  together
- get industry to build some of the hardest parts
   - data management (APL),  visualization (Google)
- get data out of institutional resources
  -  AWS
- focus on answering specific neurobiology questions with data where n>0

---

###  What is our software stack?

.footnote[https://neurodata.io/help/]

---

### What is an example use  case?

- 20 CLARITY brains, from two different conditions
- we desire to know are the two classes  different?
- and if so, how?

.center[<img src="images/CLARITY3.png" STYLE="width:50%;"/>]

---

### What is the workflow?

1. TeraStitcher
2. NDPush
3. bossDB
4. NeuroGlancer
5. NDPull
6. NDReg
7. MGC
8. LOL
8. RerF
8. graphstats

---

### TeraStitcher

- align & stitch collects of 2D images into 3D volumes
- uses  linear registration

.footnote[https://abria.github.io/TeraStitcher/]

---

### NDPush &  NDPull data to/from  cloud

- images: uint8 and  uint16
- annotations: uint32
- file formats: png, tiff, jpg, numpy, etc.
- z-slice: flexible naming conventions
- multi-channel
- pip installable*

.footnote[https://github.com/neurodata/ndpush and https://github.com/neurodata/ndpull]

---

### bossDB to store/manage data in cloud

- based on Open Connectome Project code from 2011
- ported  to AWS for scalability
- stores small cuboids
- organized into space filling curve
- random access to arbitrary cutouts
- downsamples
- experimental metadata storage
- authentication

.footnote[https://github.com/jhuapl-boss/boss]

---

### NeuroGlancer to  Visualize

.pull-left[
- Google's NeuroGlancer'
- 3D pan, zoom, &  rotate
- Multi-channel  overlays
- Select individual ROIs
- 30 minutes to load once
]

.pull-right[
<img src="images/ndviz2.jpg"  style="width: 120%;"/>
]

.footnote[https://github.com/google/neuroglancer]

---

###  NDReg to Register CLARITY to ARA

- only ~1 hr per brain
- fully automatic (no landmarks)
- works on iDisco and  other species too

.footnote[https://github.com/neurodata/ndreg]

---
class: middle

# <center>[demo](https://tinyurl.com/yco8h787)</center>

---
class: middle

### <center>pre-processing over, on to statistics</center>

---
### Multiscale Graph Correlation

- are CLARITY control brains different from conditioned brains?
- if so, how are they different?
- MGC implements universally consistent hypothesis testing
- and can reveal the geometry of the relationship
- key:  compare "correlations" between all .r[local] pairs of points
- must define a metric on CLARITY brains (eg,  distortion)

.footnote[https://github.com/neurodata/mgc]

---

###  MGC Empirically Dominates

---

###  MGC Reveals Relationship Geometry

---

### LOL for Dimensionality Reduction

Given pairs of high-dimensional X and corresponding class labels Y, find the best low-dimensional representation of X

#### Main Idea

.r[LOL]: Use the means and PCA of .r[each class separately]

.footnote[https://github.com/neurodata/lol]

---

###  LOL Empirically Dominates

---

### Random Projection Forests

Given pairs of high-dimensional X and corresponding class labels Y, find the best discriminant boundary

#### Main Idea

.r[RerF]: Find many good .r[sparse] random projections of data

.footnote[https://github.com/neurodata/RerF]

---

### Random Projection Forests Empirically Dominates

.pull-left[
- random forests  (RF) were best
- recent extensions improve
- we are even better
]

.pull-right[
<img src="images/error_histogram_benchmarks.png"  style="width: 70%;"/>
]

---
### Connectome Coding via Graph Stats

Characterizing the relationship between
the *past environment* and the *present neural connectivity*

.footnote[https://github.com/neurodata/graphstats]

---
class: middle

# <center>Discussion</center>

---
class: top, left

### Acknowledgements

<div class="container">
  <img src="faces/cep.png"/>
  <div class="centered">Carey Priebe</div>
</div>

<div class="container">
  <img src="faces/randal.jpg"/>
  <div class="centered">Randal Burns</div>
</div>

<div class="container">
  <img src="faces/mim.jpg"/>
  <div class="centered">Michael Miller</div>
</div>

<div class="container">
  <img src="faces/bcaffo.jpg"/>
  <div class="centered">Brian Caffo</div>
</div>

<div class="container">
  <img src="faces/milham.jpg"/>
  <div class="centered">Michael Milham</div>
</div>

<div class="container">
  <img src="faces/dtward.jpg"/>
  <div class="centered">Daniel Tward</div>
</div>

<div class="container">
  <img src="faces/avanti.jpg"/>
  <div class="centered">Avanti Athreya</div>
</div>

<div class="container">
  <img src="faces/vince.jpg"/>
  <div class="centered">Vince Lyzinski</div>
</div>

<div class="container">
  <img src="faces/dpmcsuss.jpg"/>
  <div class="centered">Daniel Sussman</div>
</div>

<div class="container">
  <img src="faces/yichen.jpg"/>
  <div class="centered">Yichen Qin</div>
</div>

<div class="container">
  <img src="faces/youngser.jpg"/>
  <div class="centered">Youngser Park</div>
</div>

<div class="container">
  <img src="faces/cshen.jpg"/>
  <div class="centered">Cencheng Shen</div>
</div>

<div class="container">
  <img src="faces/shangsi.jpg"/>
  <div class="centered">Shangsi Wang</div>
</div>

<div class="container">
  <img src="faces/ebridge.jpg"/>
  <div class="centered">Eric Bridgeford</div>
</div>

<div class="container">
  <img src="faces/vikram.jpg"/>
  <div class="centered">Vikram Chandrashekhar</div>
</div>

<div class="container">
  <img src="faces/tyler.jpg"/>
  <div class="centered">Tyler Tomita</div>
</div>

<div class="container">
  <img src="faces/james.jpg"/>
  <div class="centered">James Brown</div>
</div>

<div class="container">
  <img src="faces/disa.jpg"/>
  <div class="centered">Disa Mhembere</div>
</div>

<div class="container">
  <img src="faces/drishti.jpg"/>
  <div class="centered">Drishti Mannan</div>
</div>

<div class="container">
  <img src="faces/jesse.jpg"/>
  <div class="centered">Jesse Patsolic</div>
</div>

<div class="container">
  <img src="faces/falk_ben.jpg"/>
  <div class="centered">Benjamin Falk</div>
</div>

<div class="container">
  <img src="faces/kwame.jpg"/>
  <div class="centered">Kwame Kutten</div>
</div>

<div class="container">
  <img src="faces/perlman.jpg"/>
  <div class="centered">Eric Perlman</div>
</div>

---

### Questions? Now hiring!

.pull-left[
| task |  link |
| --- | --- |
| testing | [MGC](https://github.com/neurodata/mgc) |
| dim red | [LOL](https://github.com/neurodata/LOL) |
| classify | [RerF](https://github.com/neurodata/R-RerF/)  |
| graph stats |  [graphstats](https://github.com/neurodata/graphstats/)
|  registration | [ndreg](https://github.com/neurodata/ndreg) |
| email | [jovo@jhu.edu](mailto:jovo@jhu.edu) |
| web |  [neurodata.io](http://neurodata.io/) |
| startup | [gigantum.com](http://gigantum.com/) |

]

<br>