A Massively Parallel Algorithm for Cell Classification Using CUDA.

摘要

In Bioinformatics, cell classification is the act of separating human cells into different groups based on their RNA-seq expression levels. These data can be quite large, as there are about 20,000 known human genes. Even relatively small datasets (<1000 cell samples) can contains millions of values. Computations and classifications on this data force a choice: speed or thoroughness? Many scientists today choose speed, meaning they must reduce the dataset through feature selection or some similar approach. Others may use complex computing systems to achieve thoroughness, but the cost of these systems is high. NVIDIA CUDA allows the average researcher to perform cell classification without making this choice. CUDA, running on specialized GPUs, gives desktop computers the power of a large computing system and eliminates the need to reduce the complexity of the dataset.