800 publications from this institution
With the increasing throughput of modern sequencing instruments, the cost of storing and transmitting sequencing data has also increased dramatically. Although many tools have been developed to compress sequencing data, there is still a need to develop a compressor with a higher compression ratio. We present a two-step framework for compressing sequencing data in this paper. The first step is to repack original data into a binary stream, while the second step is to compress the stream with a LZMA encoder. We develop a new strategy to encode the original file into a LZMA highly compressed stream. In addition an FPGA-accelerated of LZMA was implemented to speedup the second step. As a demonstration, we present repaq as a lossless non-reference compressor of FASTQ format files. We introduced a multifile redundancy elimination method, which is very useful for compressing paired-end sequencing data. According to our test results, the compression ratio of repaq is much higher than other FASTQ compressors. For some deep sequencing data, the compression ratio of repaq can be higher than 25, almost four times of Gzip. The framework presented in this paper can also be applied to develop new tools for compressing other sequencing data. The open-source code of repaq is available at: https://github.com/OpenGene/repaq.
The phase identification and travel time picking are critical for seismic tomography, yet it will be challenging when the numbers of stations and earthquakes are huge. We here present a method to quickly obtain P and S travel times of pre-determined earthquakes from mobile dense array with the aid from long term phase records from co-located permanent stations. The records for 1 768 M ≥ 2.0 events from 2011 to 2013 recorded by 350 ChinArray stations deployed in Yunnan Province are processed with an improved AR-AIC method utilizing cumulative envelope and rectilinearity. The reference arrivals are predicted based on phase records from 88 permanent stations with similar spatial coverage, which are further refined with AR-AIC. Totally, 718 573 P picks and 512 035 S picks are obtained from mobile stations, which are 28 and 22 times of those from permanent stations, respectively. By comparing the automatic picks with manual picks from 88 permanent stations, for M ≥ 3.0 events, 81.5% of the P-pick errors are smaller than 0.5 second and 70.5% of S-pick errors are smaller than 1 second. For events with a lower magnitude, 76.5% P-pick errors fall into 0.5 second and 69.5% S-pick errors are smaller than 1 second. Moreover, the Pn and Sn phases are easily discriminated from directly P/S, indicating the necessity of combining traditional auto picking and integrating machine learning method.
