Abstract
We have used serverless AWS Lambda functions to align 640 million reads in less than 3 minutes, a speed-up of 500x over the single-threaded implementation. Using a hybrid cloud architecture and software modified to optimize disk transfers, an entire RNA sequencing workflow transforming multiplexed reads to transcript counts that originally took 29 hours can be completed in 18 minutes. This is a 100x improvement over the original single threaded implementation and 12x faster than an optimized cloud server-based implementation using 16 threads. The total cost of the analyses is $2.82 for 96 wells or 3 cents per multiplexed sample. This approach can be used for human datasets that are generated for single experiments and does not rely on processing large numbers of samples to achieve the performance gains. The workflow is publicly available under a M.I.T. license (https://github.com/BioDepot/RNA-seq-lambda).
