Enabling Cloud and Grid Powered Image Phenotyping Nirav Merchant iPlant Collaborative [email protected].

Enabling Cloud and Grid Powered Image Phenotyping

Nirav MerchantiPlant Collaborative

[email protected]

Topic Coverage

• Motivation• Key Components • Overview of BISQUE• Roadmap and future direction

Motivation• High throughput imaging is essential for enabling

genome scale phenotyping efforts• Affordable automation for image acquisition (e.g.

robotic high throughput systems) is creating vast amounts of imaging data (rapidly)

• Many laboratories have custom or commercial setup for high throughput image acquisition (but lack the comparable analysis platform)

• Super resolution microscopy and multi-channel images are pushing the boundaries of storage and computational capabilities

Image acquisition

Robotic image acquisition of root tips (Spalding Lab.)

Image Acquisition

Multiple setups recording movies for root growth (Spalding Lab.)

Motivation II• New improved algorithms and analysis routines are

being constantly published• Applying these algorithms to existing data is

challenging for biologists• Sharing and collaborating with large image data sets is

challenging• There is no common platform to try multiple

methods/algorithms on collection of images• Data management is challenging for high throughput

methods (metadata is key)• Establishing consistent protocol for image analysis is

challenging when using multiple applications/platforms

• ONE SIZE FITS ALL APPROACH DOES NOT WORK

Key iPlant infrastructure

• iPlant Data Store (iDS)• Computational Grid (HPC, HTC)• Atmosphere (Cloud Infrastructure)*• BISQUE*

iPlant Data StoreConnecting people with data and computation:

Lifecycle of DataLifecycle of Data

Transfer Storage Analysis Visualization Metadata Mark-up Search and Discover Share/Collaborate Publish

Transfer Storage Analysis Visualization Metadata Mark-up Search and Discover Share/Collaborate Publish

Why cloud ?• Standalone interactive GUI-based applications are

frequently required for analysis • GUI apps not easily to transform into web apps (or

run on grid/command line etc.)• Need to handle complex software dependencies

(e.g specific version on software/library)• Users needing full control of their software stack

(occasional sudo/super users access)• Need to share desktop/applications for

collaborative analysis (remote collaborators)

So how does it work ?

Configured VM(all required s/w)Configured VM

(all required s/w)

iPlant Data storeiPlant Data store

High B

andw

idth

Trans

fer

How does it look ?

How does it look ?

Why Bisque

• Allows algorithms developers to publish new analysis methods and make it completely web accessible with ease

• Biologists can choose from multiple analysis options for their images, overlay results to validate findings without altering original image content

• Produce interactive plots, visualization using built in API

• Share results, images , annotations with collaborators via secure link.

• Integrated with iPlant storage and computation infrastructure

Bisque features• Rich internet application (completely web based)• Draws upon features from popular large scale photo

sharing sites and high resolution aerial imagery (google maps)

• Ability to import and export over 100+ image formats, movies

• Ability to import extremely large image sets using iPlant storage infrastructure

• Can display 20Kx20K image using standard web browser• Utilizes distributed computing (connected to XSEDE) and

workflow engines (Pegasus, Condor) to scale analysis

Whole seedling-size analysisHigh resolution flat bed scanner image of seeds

Edge detection and analysis by PhytoBisque

Source: Edgar Spalding

Simple Steps for Using it

• Concept of Mini-Apps• Browse and select image (or video) • Run analysis• Overlay results and verify• Export data

PhytoBisque interface

Searching, browsing

PhytoBisque Interface

Viewing large (18Kx17k pixel image) and performing analysis on selected section

Participants

• Bisque (Univ. of California, Santa Barbara)– B. S. Manjunath– Kris Kvelikval– Dmitry Fedorov

• Phytomorph (Univ. of Wisconsin, Madison)– Edgar Spalding– Nathan Miller– Logan Johnson

Users

• Currently we have 5+ groups actively using this infrastructure

• 3 Graduate course• 2 Summer courses/workshops• 1 Pollen Network RCN• NSF ADBC Thematic Collections Network

(Yale University led)

• Main application:– http://bovary.iplantcollaborative.org

• Support:– http://ask.iplantcollaborative.org

• Project Website– http://www.iplantcollaborative.org