The MASA-CUDAlign extension is used with the MASA architecture to align DNA sequences of unrestricted size with the Smith-Waterman and Needleman-Wunsch algorithms combined with Myers-Miller. It uses the NVIDIA CUDA platform to accelerate the computation time. This extension is able to align huge DNA sequences with more than 200 million base pairs (MBP).
The current version is equivalent to CUDAlign 4.0 publicationTPDS2016, using multiple Fickett Band optimation in stage 4JCB2019.
Latest Version: masa-cudalign-4.0.2.1028.tar.gz
tar -xvzf masa-cudalign-4.0.2.1028.tar.gz
cd masa-cudalign-4.0.2.1028
./configure
make
./cudalign [options] seq1.fasta seq2.fasta
All the command line arguments can be retrieved using the --help parameter. See the wiki for a list of command line examples.
We have executed MASA-CUDAlign in many different environments. Here we will present the best results in different scenarios. We recommend to read the reference papers in order to understand the feature improvements in each test.
Test Environment (Homogeneous GPU cluster): TPDS2021
Large GPU cluster - 512 x NVidia V100.
(block pruning features in multiple GPUs using a score-share workload distribution - Figueiredo PhD thesis)
Aditional source code provided in: SLURM-Scripts, Static-MultiBP and Dynamic-MultiBP.
| Sequence 1 | Sequence 2 | Len1 | Len2 | Time | GCUPS |
|---|---|---|---|---|---|
| NC_000001.9 | NC_006468.3 | 249M | 228M | 11m | 82822 |
Test Environment (Homogeneous GPU cluster): TPDS2016
XSEDE Keeneland Cluster - 384 x NVidia Tesla M2090.
(with traceback execution/full alignment retrieval)
| Sequence 1 | Sequence 2 | Len1 | Len2 | Time | GCUPS |
|---|---|---|---|---|---|
| NC_000005.9 | NC_006472.3 | 180M | 183M | 53m8s | 10370 |
Test Environment (Homogeneous GPU cluster): CCGRID2014
Minotauro Cluster - 64 x NVidia Tesla M2090.
| Sequence 1 | Sequence 2 | Len1 | Len2 | Time | GCUPS |
|---|---|---|---|---|---|
| NC_000001.10 | NC_006468.3 | 249M | 228M | 9h09m25s | 1726.47 |
Test Environment (Heterogeneous GPU Nodes): PPOPP2014
Laico Labs - 3 Hosts - 1 x NVidia GTX 580 + 2 x NVidia GTX 680.
| Sequence 1 | Sequence 2 | Len1 | Len2 | Time | GCUPS |
|---|---|---|---|---|---|
| NC_000019.9 | NC_006486.3 | 59M | 64M | 7h29m17s | 139.60 |
| NC_000020.10 | NC_006487.3 | 63M | 62M | 7h42m08s | 140.31 |
| NC_000021.8 | NC_006488.2 | 48M | 46M | 4h27m04s | 139.63 |
| NC_000022.10 | NC_006489.3 | 51M | 50M | 5h03m00s | 140.36 |
Test Environment (Heterogeneous GPUs in Single Node): PPOPP2014
Panoramix Host - 1 x NVidia Tesla K20c + 2 x NVidia Tesla C2050.
| Sequence 1 | Sequence 2 | Len1 | Len2 | Time | GCUPS |
|---|---|---|---|---|---|
| NC_000019.9 | NC_006486.3 | 59M | 64M | 10h20m52s | 101.02 |
| NC_000020.10 | NC_006487.3 | 63M | 62M | 7h42m08s | 100.96 |
| NC_000021.8 | NC_006488.2 | 48M | 46M | 6h10m38s | 100.62 |
| NC_000022.10 | NC_006489.3 | 51M | 50M | 7h03m36s | 101.38 |
Test Environment (Single GPU): TPDS2013
NVIDIA GeForce GTX 560 Ti
| Sequence 1 | Sequence 2 | Len1 | Len2 | Time | GCUPS |
|---|---|---|---|---|---|
| CP000051.1 | CP000051.1 | 1M | 1M | 43s | 25.82 |
| BA000035.2 | BX927147.1 | 3M | 3M | 6m07s | 28.15 |
| AE016879.1 | AE017225.1 | 5M | 5M | 9m18s | 48.98 |
| NC_005027.1 | NC_003997.3 | 7M | 5M | 22m01s | 28.28 |
| NT_033779.4 | NT_037436.3 | 23M | 25M | 5h29m17s | 28.59 |
| NC_000024.9 | NC_006492.2 | 59M | 24M | 13h05m23s | 30.18 |
| BA000046.3 | NC_000021.7 | 33M | 47M | 8h26m09s | 50.70 |
MASA-CUDAlign is an open source project with public license (GPLv3). A copy of the LICENSE is maintained in this repository.
| [TPDS2021] | Parallel Fine-Grained Comparison of Long DNA Sequences in Homogeneous and Heterogeneous GPU Platforms With Pruning. TPDS 2021: 3053--3065. Marco Figueiredo, João Navarro, Edans Sandes, George Teodoro, Alba Melo |
| [JCB2019] | Using Multiple Fickett Bands to Accelerate Biological Sequence Comparisons. JCB 2019: 908-922. Gabriel Silva, Edans Sandes, George Teodoro, Alba Melo |
| [TOPC2016] | MASA: a Multi-Platform Architecture for Sequence Aligners with Block Pruning. TOPC:2(4):28. Edans Sandes, Guillermo Miranda, Xavier Martorell, Eduard Ayguadé, George Teodoro, Alba Melo. |
| [TPDS2016] | Genome Wide Alignment in GPU Cluster with Incremental Speculative Traceback. TPDS 2016. Edans Sandes, Guillermo Miranda, Alba Melo, Xavier Martorell, Eduard Ayguadé. |
| [CCGRID2014] | CUDAlign 3.0: Parallel Biological Sequence Comparison in Large GPU Clusters. CCGRID 2014:160-169. Edans Sandes, Guillermo Miranda, Alba Melo, Xavier Martorell, Eduard Ayguadé. |
| [PPOPP2014] | Fine-grain parallel megabase sequence comparison with multiple heterogeneous GPUs. PPOPP 2014:383-384. Edans Sandes, Guillermo Miranda, Alba Melo, Xavier Martorell, Eduard Ayguadé |
| [TPDS2013] | Retrieving Smith-Waterman Alignments with Optimizations for Megabase Biological Sequences using GPU. TPDS:24:5:1009-1021. Edans Sandes, Alba Melo |
| [IPDPS2011] | Smith-Waterman Alignment of Huge Sequences with GPU in Linear Space. IPDPS 2011: 1199-1211. Edans Sandes, Alba Melo |
| [PPOPP2010] | CUDAlign: using GPU to accelerate the comparison of megabase genomic sequences. PPOPP 2010: 137-146. Edans Sandes, Alba Melo |