This project implements exact alignment (no heuristics) between reads and reference graphs using CUDA, efficiently parallelizing both intra- and inter-sequence computations.
This tool computes the optimal alignment score between a sequence and a genomic variation graph, represented in CSR format. It guarantees exact results by filling the full dynamic programming matrix. The implementation is optimized for NVIDIA GPUs and scales well across both short and long reads.
To build the project:
git clone https://github.com/leonardotisato/ALIGNER-INTRA-GPU.git
cd ALIGNER-INTRA-GPU
makeTo clean the build:
make cleanYou can run the aligner directly on any input files using:
./poagpu <num_blocks> <reads_file.fa> <graph_file.gfa>Arguments:
<graph_file.gfa>: input graph in GFA format<reads_file.fa>: input reads in FASTA format<num_blocks>: number of CUDA blocks to use
This command allows you to test the aligner with any graph and read set of your choice.
Alternatively, you can test the aligner using our predefined synthetic datasets with the following Python script:
python scripts/run_poagpu.py <num_blocks> <num_vertices> <num_reads> <len_reads> [--example]Arguments:
-
<num_blocks>: number of CUDA blocks to use -
<num_vertici>: number of vertices in the graph (as indicated in the filename) -
<num_reads>: number of reads (e.g. 100, 1000, 10000 — will be auto-formatted to the find the right folder) -
<len_reads>: read length (as indicated in the filename) -
--example(optional): use example files from test/examples/ directory
This script allows you to reproduce our benchmarks or run structured tests with minimal manual setup.
src/— CUDA and C++ source filesinclude/— Header filestest/— Example input graphs and read setsdoc/— Project report
Developed by Leonardo Tisato and Gabriele Amodeo.
Supervised by Ph.D Student Mirko Coggi and Prof. Marco Domenico Santambrogio.