Search for: All records

Abstract The amount of data produced by genome sequencing experiments has been growing rapidly over the past several years, making compression important for efficient storage, transfer and analysis of the data. In recent years, nanopore sequencing technologies have seen increasing adoption since they are portable, real-time and provide long reads. However, there has been limited progress on compression of nanopore sequencing reads obtained in FASTQ files since most existing tools are either general-purpose or specialized for short read data. We present NanoSpring, a reference-free compressor for nanopore sequencing reads, relying on an approximate assembly approach. We evaluate NanoSpring on a variety of datasets including bacterial, metagenomic, plant, animal, and human whole genome data. For recently basecalled high quality nanopore datasets, NanoSpring, which focuses only on the base sequences in the FASTQ file, uses just 0.35–0.65 bits per base which is 3–6$$\times$$ $\times$lower than general purpose compressors like gzip. NanoSpring is competitive in compression ratio and compression resource usage with the state-of-the-art tool CoLoRd while being significantly faster at decompression when using multiple threads (> 4$$\times$$ $\times$faster decompression with 20 threads). NanoSpring is available on GitHub athttps://github.com/qm2/NanoSpring. 

Compression and efficient storage ofneural network (NN)parameters is critical for applications that run on resource-constrained devices. Despite the significant progress in NN model compression, there has been considerably less investigation in the actualphysicalstorage of NN parameters. Conventionally, model compression and physical storage are decoupled, as digital storage media witherror-correcting codes (ECCs)provide robust error-free storage. However, this decoupled approach is inefficient as it ignores the overparameterization present in most NNs and forces the memory device to allocate the same amount of resources to every bit of information regardless of its importance. In this work, we investigate analog memory devices as an alternative to digital media – one that naturally provides a way to add more protection for significant bits unlike its counterpart, but is noisy and may compromise the stored model’s performance if used naively. We develop a variety of robust coding strategies for NN weight storage on analog devices, and propose an approach to jointly optimize model compression and memory resource allocation. We then demonstrate the efficacy of our approach on models trained on MNIST, CIFAR-10, and ImageNet datasets for existing compression techniques. Compared to conventional error-free digital storage, our method reduces the memory footprint by up to one order of magnitude, without significantly compromising the stored model’s accuracy.