NSF NeuroNex Workshop (3DEM)

name:opening

<img src="images/neurodata_blue.png" style="height:150px;float:right;"/>
### NeuroData: Embracing Open Source for Big Data Neuroscience

Eric Perlman

.foot[perlman@jhu.edu | <http://neurodata.io/talks/> ]

---
class: center

### <center>What is NeuroData?</center>
<img src="images/neurodata_venn.png" STYLE="width:50%;"/>

---
### Origin: Hosting public data (2011-)
<img src="images/website_datapage.png" style="width: 100%;"/>

.foot[<http://neurodata.io/data/>]

---
### ... but, it's hard.

- Don’t work alone!
  - In 2011, Open Connectome Project set out to revolutionize data management for EM neuroscience with storage and analytics services on our cluster
  - And we have had mixed results...
  - ...but excited to have seeded a second generation
---
### The New Role of neurodata.io...

Shift from "build your own" to "see what's available; supplement where necessary."
- To facilitate a collaborative, open-source, cloud computing ecosystem for end-to-end neuroscience
- Collaborative: multiple organizations and communities developing and using the same tools
- End-to-end: data management from microscope to publication
- Visualize data at all stages
- Reproducible, extensible analysis

---

###  What is our software stack?

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

---
### A brief history of neurodata spatial stores...

- (2011-2013) Open-Connectome Project (OCP)
  - First version of a spatial data store for volumetric cutouts and annotation
- (2013-2016) NeurodataStore (ndstore)
  - Clustered, scale-out reimplementation of OCP on hosted NoSQL clusters
  - Migration to the cloud in 2016
- (2014-present) IARPA MICrONS program drives Boss development
  - JHU/APL builds and deploys for MICrONS performers
- (2016-present) neurodata.io adopts BossDB for open-science
   - Parallel deployment of cloud services, independently managed
   - Contribute features back through pull requests

---

### BossDB Features & Benefits

- ported to AWS for scalability
- random access to arbitrary cutouts
- downsamples
- experimental metadata storage (e.g. JSON)
- authentication
- spatial queries on annotations

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

---

### NDPush & NDPull data to/from cloud

- images: uint8 and uint16
- annotations: uint64
- file formats: png, tiff, jpg, numpy, etc.
- z-slice: flexible naming conventions
- multi-channel
- pip installable

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

---

### Is a central (active) data store the right choice?

...not always.
---
### Other storage options

- hdf5
- n5 & Zarr provide chunked N-dimensional data stores
  - same api for local or remote and parallel writes
- neuroglander's precomputed chunk & compressed segmentation format

- We are also exploring "Boss direct io", bypassing the service layer

.foot[https://github.com/saalfeldlab/n5/] 
.foot[https://github.com/zarr-developers/zarr/]

---
### NeuroGlancer for Visualization

.pull-left[
Visualization with NeuroGlancer at all stages!
- Acquistion
- Alignment
- Analysis
]

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

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

---

### Modifications to NeuroGlancer
- Boss support
  - Including authentication
- Color selection
  - Color picker & min/max values to generate shaders
  - Used heavily on LM datasets
- Other ways to interact with state (export, load remote)
  - Static sharing of views w/o python backend.
- Extra visualizations (e.g. render pair matches)
- Ontology info (e.g., display the region name from the ARA)

---

### Render Tools & services

- Web-based version of TrakEM2's XML datastore
- Stores references to raw data & transformations
- RESTful interfaces to load & update data
- Dynamic rendering of data
- Extended to support cloud data sources (e.g., raw images stored in an AWS)

.foot[https://github.com/saalfeldlab/render/]

---
### Alignment verification...
<img src="images/synaptomes_pointmatch.png" style="width:100%;"/>

---
### Correlated EM-Light
<img src="images/synaptomes_em_light.png" style="width:100%;"/>

---
### Light sheet registration (ndreg)

- LDDMM
- Fully automatic (no landmarks)
- Works on iDisco, Clarity, and multiple species...

---
### Other tools ...

- Image analysis
- Statistical machine learning (LOL, RerF, FlashX, knor, ...)

.foot[https://neurodata.io/tools/]

---
### Acknowledgements

- neurodata: Randal Burns, Joshua Vogelstein & Ben Falk
- viz: Alex Baden & Jeremy Maitin-Shepard (Google)
- ndreg: Vikram Chandrashekhar & Daniel Tward
- synaptomes: Jesse Patsolic, Forrest Collman (AIBS), Sharmi Seshamani (AIBS), Stephen Smith (AIBS)
- render: Stephan Saalfeld (Janelia) & Eric Trautman (Janelia)

---
class:center