This is a GPU-based implementation of a sliding window matrix-scoring method in Python.

It speeds up scoring of large amounts of sequence data, useful for larger datasets like metagenomes.

In this repository, you will find:

• gpu_pssm.py: Contains the main scoring function, score_sequence(). See the function definition for further documentation and usage examples.
• cuda_benchmark.py: Benchmarking code used to help evaluate the best parameters to use with your CUDA-capable card.
• score_metagenome.py: Utilizes the GPU approach to scoring a large dataset (the MetaHit database) against a motif of LexA binding sites.

The dependencies for these scripts are:

• NumPy
• NumbaPro: This was chosen as it is NVIDIA's endorsed Python implementation of CUDA. It has good support and is in active development, plus the license is free to academics (anyone with a .edu email address) so it can be freely used for research. To get NumbaPro, first install Anaconda and then follow the instructions on the NumbaPro installation page. It's a surprisingly painless install!
• CUDA Toolkit: NVIDIA's CUDA toolkit which is necessary to provide functionality for GPU programming.
• A CUDA-capable GPU with compute-capability of 2.0 or higher.

This implementation is based on CUDA 5.5 and hopefully will get upgraded when CUDA 6 comes out with Unified Memory.