A Vision for Managing Big Data @ UC Davis A Data Science Institute

advertisement
Big Data
Why it matters
Patrice KOEHL
Department of Computer Science
Genome Center
UC Davis
The three I’s of Big Data
Big Data is:
- Ill-defined (what is it?)
- Immediate (we need to do something about it now)
- Intimidating (what if we don’t)
(loosely adapted from Forbes)
Big Data: Volume
Byte
Kilobyte Megabyte
KB
MB
1000 bytes 1000 KB
Gigabyte Terabyte
GB
TB
Petabyte Exabyte
PB
EB
1000 MB 1000 GB 1000 TB 1000 PB
Zettabyte Yottabyte
ZB
YB
1000 ZB
1000YB
Big Data: Volume
One page One song One movie 6 million 55 storeys Data
of text
books
of DVD
up to
2003
5 MB
30KB
5 GB
1 TB
1 PB
Data
in 2011
1.8
ZB
NSA
data center
1 YB
5 EB
Byte
Kilobyte Megabyte
KB
MB
1000 bytes 1000 KB
Gigabyte Terabyte
GB
TB
Petabyte Exabyte
PB
EB
1000 MB 1000 GB 1000 TB 1000 PB
Zettabyte Yottabyte
ZB
YB
1000 ZB
1000YB
Big Data: Volume
One page One song One movie 6 million 55 storeys Data
of text
books
of DVD
up to
2003
5 MB
30KB
5 GB
1 TB
1 PB
Data
in 2011
1.8
ZB
NSA
data center
1 YB
5 EB
Byte
Kilobyte Megabyte
KB
MB
1000 bytes 1000 KB
1s
20 mins
11 days
Gigabyte Terabyte
GB
TB
Petabyte Exabyte
PB
EB
1000 MB 1000 GB 1000 TB 1000 PB
30 years
Zettabyte Yottabyte
ZB
YB
1000 ZB
300
30 million 30 billion ….
centuries years
years
1000YB
Big Data: Volume, Velocity
One minute in the digital world
(Intel, 2013)
204 million
640 TB
e-mails
sent
IP data
transferred
50 GB
of data
generated
at the Large
Hadron Collider
3+ million
searches
launched
6 million
users
connected
30 hours
videos
uploaded
1.3 million
videos
viewed
Big Data: Volume, Velocity, Variety
Numbers
text
Images
sound
Big Data: Challenges
 Volume and Velocity
 Variety
 Structured, Unstructured….
 Images, Sound, Numbers, Tables,…
 Security
 Reliability, Integrity, Validity
Big Data: Challenges
Large N:
“Any dataset that is collected by a scientist whose data
collection skills are far superior to her analysis skills”
Computing issues:
 Data transfer
 Scalability of algorithms
 Memory limitations
 Distributed computing
Big Data: Challenges
Vizualization issues:
The black screen
problem
(Matloff, 2013)
Big Data: Challenges
Rule of thumb: N/P > 5….what if it does not
hold anymore?
Large P, “small” N:
 Curse of dimensionality
(all data points seem equidistant)
 Non linearity
 Dimension reduction
Big Data: Challenges and Opportunities
 Fourth Paradigm: data driven science
Basic
Data
Translational
Knowledge
Societal Benefit
 Holistic approaches to major research efforts
 New paradigms in computing
 Digital Humanities
Big Data: Enabling Dreams
 Understanding the physics of “Dark Energy”
 How the brain works: from neurons to cognition
 A holistic view of natural ecosystems
 Understanding climate changes
 From genotype to phenotype
 Precision medicine
 Big Humanities
 ….
Big Data Dreams: Genomics
Big Data Dreams: Genomics
Genomics: Sequencing costs
Cost per Mbase
$1,000.00
$100.00
$10.00
$1.00
$0.10
$0.01
$100,000,000
Cost per Human Genome
$10,000.00
$10,000,000
$1,000,000
$100,000
$10,000
$1,000
$100
http://www.genome.gov
Genomics: Game changing technologies
Illumina HiSeq 2000
Capable of 600 Gb per run -> 1,000+ Gb
55 Gb/day
6 billion paired-end reads
<$4,000 per human/plant genome
<$200 per transcriptome
Multiplex 384 pathogen isolates/lane
 $10 (+ $50 library construction)/isolate
Challenges: Library preparation &
data analysis
Gary Schroth (Illumina): “A single lab with one HiSeq is able to generate
more sequences than was in GenBank in 2009, every four days”.
Genomics @ UC Davis
Massively parallel DNA sequencing
2 Illumina Genome Analyzers
1 Illumina Hiseq 2000, 2 Miseq
1 Roche 454 Junior
1 Pacific Biosystems RS
GoldenGate SNP genotyping
iScan, BeadArray & BeadExpress
Cancer Genomics: Molecular Diagnostics
Genomics: actual costs
“A single lab with one HiSeq
is able to generate more
sequences than was in
GenBank in 2009, every four
days.”
Gary Schroth (Illumina)
Genomics: actual costs
Assembling 22GB conifer genome:
“A single lab with one HiSeq
is able to generate more
sequences than was in
GenBank in 2009, every four
days.”
Gary Schroth (Illumina)
Data:
-16 billion pair reads (100 bases)
Processing:
-10 days for error correction
-11 days for assembling “super-reads”
-60 days to build contigs/scaffold
-8 days to fill in gaps
http://www.homolog.us/blogs/2013/05/11/
steven-salzberg-at-bog13-assembling-22gb-conifer-genome/
Social Consequences of Commodity Sequencing
 The danger of misuse
predict sensitivities to various industrial or environmental agents
discrimination by employers?
 The impact of information that is likely to be incomplete
an indication of a 25 percent increase in the risk of cancer?
 Reversal of knowledge paradigm
 Are the "products" of the Human Genome Project to be
patented and commercialized?
Myriad genetics and BRCA1/2
 How to educate about genetic research and its
implications?
Social Consequences of Commodity Sequencing
Social Consequences of Commodity Sequencing
How to Approach Big Data
Download