2. 2OUR GOAL
Speed up the identification of
DNA variation while lowering
the power consumption
3. 3
HARDWARE AND
GENOMICS
GENOMICS ANALYSIS
INVOLVES
• Big amount of data
• Both sequential and parallel
processes
• Hidden Markov Model
(6 times faster on FPGA then on GPU)
http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=1287823
4. ALGORITHMS’ PIPELINE: XHMM
Mean per-target
coverage using GATK
Filter out «extreme»
target and simple
Filter target and samples
and calculate z-score
Mean center each
target and run PCA
HMM to merge targets
and call per-sample CNVs
5. ALGORITHMS’ PIPELINE: CLAMMS
• High Scalability
• Both parallel and sequential
processes
• k-Nearest Neighbors
(8 times less power consuming on
FPGA then on GPU)
• Depth Of Coverage with GATK
• HMM
https://www.ncbi.nlm.nih.gov/pubmed/26382196
http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7160066