SCA-NGS: Secure compression algorithm for next generation sequencing data using genetic operators and block sorting

Abstract

Recent advancements in sequencing methods have led to significant increase in sequencing data. Increase in sequencing data leads to research challenges such as storage, transfer, processing, etc. data compression techniques have been opted to cope with the storage of these data. There have been good achievements in compression ratio and execution time. This fast-paced advancement has raised major concerns about the security of data. Confidentiality, integrity, authenticity of data needs to be ensured. This paper presents a novel lossless reference-free algorithm that focuses on data compression along with encryption to achieve security in addition to other parameters. The proposed algorithm uses preprocessing of data before applying general-purpose compression library. Genetic algorithm is used to encrypt the data. The technique is validated with experimental results on benchmark datasets. Comparative analysis with state-of-the-art techniques is presented. The results show that the proposed method achieves better results in comparison to existing methods.

Introduction

Recent advancements in sequencing methods have led to tremendous increase in sequencing data due to which many problems arise such as memory, storage, transfer, processing, confidentiality, etc. ¹ In recent years, the rate of generating biological data has increased significantly. In early times sequencing cost was high with low speed of data generation. Modern machines generate data at a high speed at reduced cost. ² The 1000 genome project has produced a huge amount of data, that is, the generated data in the first 6 months of the project is more than the total data of NCBI GenBank data in 21 years before the project. ³ The Beijing Institute of genomics produced 30 petabytes of raw data per year. ⁴ The increase has left behind the computing power of modern computers. ² This huge increase has motivated researchers to develop compression tools to handle the huge amount of data. There have been good achievements in compression ratio and execution time.

Data compression is a technique to reduce the size of data and address issues related to storage and transfer. There has been extensive development in compression tools for genomic data. Different techniques have been formulated for the compression of biological sequences. Techniques for genomic data take FASTA files as input and process data with both lossy and lossless techniques. For NGS data FASTQ and other formats, files are taken as input and processed with both lossy and lossless compression. Different generalized methods have been used for data compression. One of these methods is the statistical method. The statistical approach is a technique widely used for compression. The technique is based on encoding one symbol at a time and works on the probability of occurrence of symbols. The general methods of this class are Huffman and arithmetic coding. Another technique known as the Lempel-Ziv method is based on the classic creation of dictionaries. ⁵ Dictionary consists of repeated substrings in data. The specialty of Lempel-Ziv is that it records the whole genomic dataset in dictionary rather than only a reference genome. In data compression methods, the permutation of data can be utilized to improve compression, for example, the assembled genomic data. This method yields better results in terms of compression ratio. Examples of this method include table compression and Burrow Wheeler Transform (BWT).

Different approaches have been opted to compress biological data. Due to the unique characteristics of biological data, general-purpose compression techniques are not suitable to compress these data. Taking this motivation into consideration, specialized compressors have been developed. These include reference-free or reverential compression tools. Some approaches use lossy compression whereas others use lossless compression. Due to high transfer costs, the data need to be stored in the cloud. The data in cloud environment need some security measures. Despite the rich literature for the data compression technique, general-purpose compression methods are still employed for compressing genome data. Many of the repositories use general-purpose compression tools. Specialized compression needs to target industrial trends of compression and develop tools that can replace general-purpose compression techniques as well as consider data security.

The fast-paced advancements have raised major concerns for the security of data. Large volumes of data have raised concerns for secure storage, privacy, and accessibility of data. ⁶ Encryption along with compression is the solution to address these issues. Cryptographic schemes can be applied for the security and confidential accessibility of data. ⁶ The use of cloud computing and blockchain (for bioinformatics applications) opens new challenges for compression tools. Encryption of compressed datasets is a challenging task that needs to address security along with other parameters. General-purpose compression and encryption tools can directly be applied to genome data, but these tools cannot take advantage of the characteristics of the genomic data. The characteristics include small alphabet size, repetitive nature of the occurrence of the patterns, palindromes, etc. Most of the existing solutions do not consider security as a parameter. Only a few tools have addressed the problem of data security. Adding this parameter to the existing method will affect execution time and other parameters.

This article presents a novel compression and encryption algorithm for NGS data. The framework utilizes general-purpose compression library for compression of quality scores and other components of the data. Bases are pre-processed before applying general-purpose compressor. The resulting data are encrypted using Genetic Algorithm (GA). Crossover and mutation phases of GA are used to encrypt the data with a given key. Comparative experimental results analysis on benchmark datasets shows the effectiveness of the proposed method. The paper is organized as follows. Section 2 presents related work followed by materials and methods in section 3. Section 4 presents the results and discussion and finally, section V concludes the article.

Literature review

Sequence data compression techniques can be either genomic data compressors or NGS data compressors. The main difference between genomic data and NGS data is the presence of quality scores in NGS data. Like genomic sequences, NGS data also consists of headers and bases. The quality score has a broad range of alphabets that makes standard compression tools and specialized tools for genomic data inappropriate for this type of data. Both types of compression tools consist of referential or reference-free methods. In referential mode, a reference sequence is used to compress the target sequence. In reference-free mode, no reference sequence is used. This section presents compression techniques for NGS datasets. The discussion covers reference-free compression tools. Readers are referred to reviews^2,4,7 on sequence data compression for further details on different methods and categories.

Literature consists of techniques based on various methods. NAF is a tool and a format for storing compressed DNA sequences. ⁸ The tool supports multiple formats for compression. The algorithm works in lossless reference-free mode. Like many other techniques for DNA sequence data compression, NAF also splits sequences into different streams. FASTQ files are split into bases, quality scores, and headers. Each component is processed separately. First, sequences are concatenated and then converted into 4-bits encoding. In the end, general-purpose compression library is used to further compressed the data. The main advantage of NAF is the fast execution time in both compression and decompression. The tool also consumes less memory in compression. Another compression tool named minicom is proposed for reads compression. ⁹ Reads are indexed with k-minimizers and sub-groups are created based on the similarity of the minimizers. In the next step, contigs are constructed in each group which are converted to larger contigs on the basis of minimizer-index suffix-prefix overlap similarity. The process is repeated until all contigs are merged. Comparative experimental results are shown to validate the performance of minicom. Another technique considers data compression and encryption as essential parameters for low-cost storage and transmission. ¹⁰ The proposed algorithm uses reverse palindromes, genetic palindrome, and sub-string for compression. Substrings of different lengths are replaced with ASCII values. In the end, encryption is applied over the compressed to ensure the security of the data. Comparative experimental results are presented to validate the proposed method. FaStore is another tool for compressing raw sequencing data. ¹¹ The algorithm works in lossless reference-free mode. The proposed method exploits the redundancy in reads to achieve better compression. The procedure includes read clustering, optional reads re-clustering, and compression stages. Each stage has further steps to achieve the desired goals. The tool has additional options for users to discard any portion of data, that is, headers or quality scores to reduce the generated file size. The input sequences are split into different streams and specialized compressors are used to compress each stream. The algorithm has also the ability to preserve the pairing information between reads. Comparative experimental results are presented to validate the performance of the proposed method. Researchers have also worked to exploit inter-similarities between sequences to achieve better compression gains. ¹² The method is based on clustering of data into similar sub-groups and applies group by group compression. First, the method detects lexicographically the smallest k-mers in each read. These k-mers are used as features and the frequencies of k-mers are used as features values to transform the datasets into feature vectors. Similar datasets are found and merged with unsupervised clustering algorithms. Experimental results are presented to validate the proposed method. Another algorithm LFQC ¹³ applies Huffman coding for quality score compression. The algorithm divides quality score data into several chunks. Each chunk is encoded individually. The algorithm avoids successive # symbols to reduce quality score alphabets. The data are compressed with a context mixing algorithm. ¹⁴ A similar procedure is used to compress bases. FQSqueezer ¹⁵ is based on partial prediction and dynamic Markov-coder algorithms. FQSqueezer compresses single-end and pair-end reads of variable length. The algorithm uses previously available techniques for compression, that is, prediction by partial matching and Dynamic Markov Coder. These techniques have been improved for organizing large dictionaries, estimation of sequencing errors, reordering reads, and sharing substrings among reads. The algorithm performs better than other compared methods and shows a significant gain in compression ratio. The main drawbacks are huge memory acquisition and long execution time. Accelerated implementation of FQSqueezer is also available. ¹⁶ The existing method runs on multi-core CPUs for multi-threading. The accelerated version uses GPUs for this purpose. The objective is to improve the performance. Improvement gain in performance is validated with experimental validations. PgRC technique is based on extracting the shortest identical superstring among similar reads. ¹⁷ PgRC shows better compression ratio than other compared techniques. The technique also has the advantage of decompression speed over the competitors. The algorithm works on approximating the shortest common superstring. Firstly, reads are partitioned based on quality and number of symbols. Reads that relate without errors are known as high-quality reads and reads that are matched including errors are known as low-quality reads. Low-quality reads that cannot be mapped wholly are partially matched with the relevant pseudo-genome area. The tools can be used to compress reads in FASTQ format. Cryfa ⁶ is proposed to secure many formats of genomic data such as FASTA, FASTQ, VCF, SAM, and BAM. The algorithm not only applies encryption to data but also compresses the data. Cryfa uses advanced encryption methods that have a key shuffling mechanism to secure data. The method is faster in compression speed than other algorithms. To encrypt and compress data are split into three streams, that is, headers, bases, and quality scores. The separate data streams are preprocessed and transformed into ASCII characters. Then a key is generated to shuffle the content. Finally, AES encryption is used to secure the data. The main advantage is low memory consumption with faster execution speed. However, the compression ratio is low compared to other methods. MZPAQ ¹⁸ is developed by combining previously developed techniques, that is MFCompress, ¹⁹ ZPAQ. ¹⁴ The technique divides data into three streams and uses strong context-mixing algorithm. The procedure then generates output in the form of single binary file. The compression process utilizes MFCompress for identifiers, bases, and ZPAQ for quality scores compression. SPRING ²⁰ is another tool proposed that exhibits different modes of compression. The tool has different features and modes that carry out the alignment of reads, lossy quality value compression, lossless large read compression, and frequent access, etc. The lossy mode records information such as alignment, assembly, variant calling. Reads can be effectively retrieved through lossy mode. The tool has support for variable length reads, random access, and support for high coverage datasets. The performance of the proposed method is evaluated with extensive experimental evaluation. The algorithm proposed in Chandak et al. ²¹ compresses both unaligned and aligned reads. In both scenarios, the algorithm achieves higher compression ratio over general-purpose compression algorithms and shows improvement gain. Reads are first preprocessed to remove non-ACGT values and are stored in separate files. Reads are ordered based on the location in the genome through hash-based substrings. The algorithm has three phases, that is, reordering, encoding, and compression. Reads are reordered by mapping their loci in the genome. After reordering, reads are encoded to remove repetition and the parameters are saved in separate files. Finally, reads are compressed through Lempel-Ziv and BWT compressors. LFastqC ²² compression tool splits each read into three data streams and compresses the streams using two compression tools, that is, lpaq8 ²³ and MFCompress. ¹⁹ Lpaq8 utilizes context-mixing algorithm. Context mixing uses prediction of weighted combination of estimates from different models. In LFastqC, MFCompress is used for DNA compression. Reads undergo minimal preprocessing before compression. The compression ratio is better than some general-purpose compressors and specialized algorithms. Ipaq8 is used to compress quality scores. FCompress ²⁴ is another tool that splits FASTQ sequence files into three separate streams. Bases are compressed with the dictionary method by taking segments of four bases. In the next phase, 7z compressor is used to compress the values obtained in the first phase. Headers are also compressed with 7z compressor. The quality score is compressed with Huffman coding. Huffman coding is applied in blocks to achieve better compression results. The method is evaluated with experimental results. GTZ is another approach for FASTQ data compression. ²⁵ The technique is also capable of transmitting compressed data to the cloud server. The approach splits FASTQ files into different streams and uses context modeling for estimating probabilities of characteristics of the input data and arithmetic coding for compression. Results on benchmark datasets are presented to validate the proposed technique. Another compression technique is proposed in Fan et al. ²⁶ The technique integrates FM-Index and complementary context models for compression. FM-Index is used to find exact matches between the two sequences. For mismatches, complementary context models are used. The technique is evaluated with benchmark datasets. A compression and encryption tool named GP²R ²⁷ processes the data in two tiers in lossless reference-free mode. In the first tier, the technique searches for substrings using exact genetic palindrome, palindrome, reverse. A library is created using corresponding ASCII characters. In the second tier, the modified RSA technique is used for encryption.

Material and methods

This section presents the details of the proposed algorithm. The methodology consists of the steps shown in Figure 1. In the first step, the proposed method splits data into different streams, that is, header, bases, and quality scores. Bases are preprocessed by splitting into combinations of four bases. ASCII values of each combination are processed further. In the next phase, each component is converted to binary, and GA-based encryption is used for encryption. In the final phase, headers, bases, and quality scores are compressed with general-purpose compression library. ²⁸

Figure 1.

Architecture of SCA-NGS.

Compression and encryption

DNA sequences are taken as the input string. First, all non-ACGT characters are removed from the input string. All lines containing Ns are transferred to separate file and their positions are recorded. This input sequence is further divided into chunks of four. We can make a dictionary of 256 possible combinations of four letters. The position of each chunk in the array is observed. The positions are taken as integer values. To speed up the process, binary search is used to search the desired position in the array. This strategy yields compression ratio equal to 2 bits per base compression. The integer values are then converted to equivalent ASCII values and converted to binary. These values are transferred to the encryption module and then to the BSC library for further compression. For decompression purpose, each binary value is transformed back to ASCII value where it is replaced by four characters.

To compress other components, each stream is encrypted and processed with BSC compression library. BSC is a reference-free program that uses block sorting method for data compression. The method uses parallel multi-threading approach to compress large partitions of data. Memory usage depends on the number of blocks being processed at a given time. Blocks can be of varied sizes. Algorithm 1 shows the steps in the compression process of the proposed method. Algorithm 1 shows the steps involved in compression.

Algorithm 1. Compression procedure

begin read line l from FASTQ file for each base line b   ifb contains non-AGCT characters    store b in separate file along with line number    process b with bsc library   else   initialize pattern array A (Array of 256 possible combinations of four DNA bases)   end if  for counter i set to 0   split b into segment s of length 4   search A for s (use binary search to find the index of s in A)   store index of si   increment i by 4  end for  call encryption procedure (algorithm 2)  compress the resulting data with bsc library  end if end for end

The proposed algorithm uses symmetric key encryption scheme based on GA, that is, crossover and mutation operators are used to encrypt the data. ²⁹ The procedure is shown in Figure 2. Algorithm 2 shows the steps involved. All components, that is, DNA, header, and quality score data are processed similarly. In the crossover phase, both the key and text are converted to binary. A random point is selected as crossover point. The point is selected between 1 and 8 as the text and key are represented in 8 bits. This point is stored along with the key for decryption. To encrypt the text, the XOR of the selected bit in each byte of key and text is taken and the bit in the text is replaced with the new bit. Changing 1 bit in a character changes the character as well and thus remains unknown. In the mutation operator, another point is selected as the mutation point. This point is different from the crossover point in the range of 1 and 8. The selected bit in each byte of the data is flipped. The mutation point is also saved with the key to be used for decryption.

Figure 2.

An illustrative example of the encryption process. In the evaluation of the proposed algorithm, the key length was kept equal to the length of the sequence line.

Algorithm 2. Encryption Procedure

begin Read line l from the given sequence. Generate key k of equal length of l. bl= l converted to binary (8 bits) bk= k converted to binary (8 bits) generate random crossover point between 1 and 8. generate random mutation point between 1 and 8. (different from crossover point) for i=0 to length of bl  ifbiti equals to crossover point   take XOR of the biti in bk and bl.   replace biti in bl with new value.  end if end for for i=0 to length of bl  ifbiti equals to mutation point   flip biti in bl  end if end for end

Decompression and decryption

All files generated during compression are decompressed first with general-purpose compression library, that is, BSC. After decompression with BSC, decryption is done with the help of key and random points for crossover and mutation generated during the encryption process. In the next step, DNA is decoded as ASCII values, segments of four bases are retrieved from the array and finally, all files are combined to regenerate the original FASTQ file. Figure 3 shows the steps involved in the decompression process. In the decryption phase, the encryption process is reversed. First, the mutation is performed by flipping the random bit in each byte of the encrypted text. In the next phase, that is, crossover the XOR of the random bit (selected in encryption phase) of each byte of the text and key is taken and the bit in data is updated accordingly to recover the plain data.

Figure 3.

The process of decompression.

Results and discussion

This section presents the experimental results of the proposed method in comparison to other state-of-the-art techniques in the literature. Experiments are performed on NGS datasets to validate the performance of the proposed SCA-NGS. The datasets are used to validate many compression tools and review articles for comparing compression programs.^2,7,24,30 Datasets used in experiments are publicly available on the NCBI website. The datasets are related to different species. (Human, plants, worm, fungus, and bacteria) of varying sizes (1.7–64.18 GB). Datasets are generated with different platforms (Illumina, 454, Solid, and Ion Torrent). The varying size of datasets acquired from multiple platforms helps in the careful evaluation of experimental results. Details related to the dataset are shown in Table 1.

Table 1.

NGS datasets used for experiments.

Datasets	Species	Number of reads	File size (MBs)
SRR489793	C. elegans	56,851,258	13132.48
SRR801793	L. pneumophila	10,812,922	2818.11
ERR022075	E. coli	45,440,200	11253.16
SRR003177	Homo sapiens	1,504,571	1672.78
SRR125858	Homo sapiens	124,815,011	52172.64
SRR935126	A. Thaliana	49,719,116	10039.24
SRR611141	Homo sapiens	4,853,655	1799.86
SRR400039	Homo sapiens	124,331,027	65723.77

Compression programs for FASTQ format are selected based on the type of compression method used, streams of data targeted for compression, and requirements of resources. Some programs are designed only for read compression whereas some programs use lossy compression. Programs selected for comparison with SCA-NGS are lossless reference-free tools designed for FASTQ file compression. Specialized compression tools used for comparison include SPRING, NAF, and FaStore. General-purpose compression tools include 7z, Bzip2, and gzip. These tools are used to make baseline for comparison. Recent tools that use encryption along with compression are used for comparison of encryption and compression parameters of SCA-NGS.

First compression parameters are compared with compression techniques using different parameters followed by compression and encryption results. Comparative results in terms of compression and decompression time, compression and decompression memory, compression ratio, compression and encryption time, and decompression and decryption time are presented. Table 2 shows the description of various parameters used for comparison.

Table 2.

Description of the parameters used for comparison. The compression ratio is calculated with equation (1).

Parameters	Description
Compression ratio	The ratio between the compressed and decompressed file size is calculated with equation (1).
Compression time	The time required to compress a file is known as compression time.
Compression memory	Peak memory required by a compressor during compression.
Decompression time	The time required to decompress a file is known as decompression Time.
Decompression memory	Peak memory required by a compressor during decompression.

All programs were executed on a computer having Intel Core i5 2.4 GHz processor, equipped with 16 GB RAM, and running Ubuntu 16.04 operating system. Each tool or program was compiled and run with respective compilers. Programs were executed with default options. SPRING and NAF use lossless mode as the default option whereas FaStore was executed with lossless option in compression mode.

$$\mathrm{Compression}\mathrm{Ratio}=\frac{\mathrm{Uncompressed}\mathrm{size}}{\mathrm{Compressed}\mathrm{size}}$$ (1)

Table 3 shows the comparative results of the proposed algorithm with other techniques. The results for various datasets show that the proposed algorithm produces better results as compared to other algorithms. SCA-NGS shows moderate results in terms of memory consumption for all datasets. General-purpose compression tools score best in terms of memory usage among other algorithms. The proposed algorithm yields better compression ratio with low execution time on many datasets as compared to specialized compression tools. In specialized compressors, NAF has smaller execution time in comparison to other tools with the cost of low compression ratio. NAF also has the advantage of low memory consumption in compression however in decompression memory consumption is high. SPRING yields compression ratio closer to SCA-NGS at the cost of higher execution time. Memory consumption of SPRING remains high for both compression and decompression.

Table 3.

Comparative experimental results of the SCA-NGS with other methods on NGS datasets. CTime and DTime refer to compression and decompression times in seconds. CMem and DMem refer to compression and decompression memory in MBs. CRatio refers to compression ratio. We could not decompress the dataset SRR400039 with NAF due to high memory requirements.

Programs	CTime	CMem	DTime	DMem	CRatio
SRR801793 (2818.11)
SCA-NGS	110	1148	58	1331	5.08
SPRING	386	3386	45	3686	5.0
NAF	30	486	28	1728	3.66
FaStore	673	872	93	736	4.69
7z	436	6.6	34	0.55	3.24
Gzip	282	0.37	49	0.18	3.05
Bzip2	439	6.4	139	3.6	3.73
ERR022075 (11253.16)
SCA-NGS	401	1311	305	1528	5.48
SPRING	1847	3548	153	5529	5.22
NAF	126	16	104	2356	3.78
FaStore	2174	1847	428	1132	4.87
7z	2378	6.6	200	0.55	3.39
Gzip	896	0.37	231	0.18	3.15
Bzip2	1002	6.4	457	3.6	3.84
SRR125858 (52172.64)
SCA-NGS	1741	1638	1531	2132	5.82
SPRING	4671	4104	519	6310	5.8
NAF	759	17	484	13643	4.24
FaStore	3856	2922	1121	1786	5.47
7z	10456	6.6	1272	0.55	3.65
Gzip	3814	0.37	1277	0.18	3.4
Bzip2	4738	6.4	3260	3.6	4.1
SRR611141 (1799.86)
SCA-NGS	44	948	36	1142	4.03
SPRING	319	2765	45	2663	3.7
NAF	32	17	13	444	2.81
FaStore	312	1245	57	1102	3.51
7z	169	6.6	19	0.55	2.62
Gzip	144	0.37	21	0.18	2.48
Bzip2	138	6.4	119	3.6	3.02
SRR489793 (13132.48)
SCA-NGS	548	1536	490	1562	4.65
SPRING	1573	3174	1354	3438	4.27
NAF	218	16	130	2765	3.53
FaStore	1272	1844	224	1636	4.21
7z	2421	6.6	236	0.55	3.18
Gzip	1027	0.37	297	0.18	2.97
Bzip2	1252	6.4	849	3.6	3.6
SRR935126 (10039.24)
SCA-NGS	282	1126	193	1433	5.48
SPRING	1269	2532	847	2728	5.22
NAF	107	16	73	2662	4.32
FaStore	871	1445	398	1136	5.16
7z	2312	6.6	258	0.55	3.72
Gzip	755	0.37	175	0.18	3.42
Bzip2	936	6.4	592	3.6	4.19
SRR003177 (1672.78)
SCA-NGS	37	1638	26	1532	5.12
SPRING	246	2472	118	2532	4.97
NAF	22	16	13	378	3.48
FaStore	169	1002	61	856	4.79
7z	336	6.6	19	0.55	3.3
Gzip	129	0.37	18	0.18	3.03
Bzip2	135	6.4	86	3.6	3.73
SRR400039 (65723.77)
SCA-NGS	2640	1556	2580	1752	4.19
SPRING	4618	3365	3112	3722	4.14
NAF	911	17	–	–	3.14
FaStore	5416	2146	1823	1540	3.94
7z	12255	6.6	1316	0.55	3.15
Gzip	4498	0.37	1175	0.18	2.96
Bzip2	6078	6.4	4137	3.6	3.64

Note: Bold entries show best results for each parameter.

For dataset SRR801793 SCA-NGS achieved percent improvement gain of 1.5, 28.4, and 10.65 in terms of compression ratio over SPRING, NAF, and FaStore, respectively. In terms of execution time, NAF is faster than the other methods. The proposed algorithm achieves 71.5, and 83.6 percent improvement gain over SPRING and FaStore.

For the ERR022075 dataset, the proposed algorithm yields better compression ratio as compared to other algorithms. SPRING remains closer to the proposed algorithm. However, the proposed algorithm has the advantage in compression and decompression time. The percent improvement gains of the proposed algorithm over SPRING and FaStore in terms of execution time is 78.2 and 81.5, respectively. NAF leads all the compared methods in execution time. In terms of compression ratio, the proposed algorithm achieves percent improvement gain of 4.74, 23.37, and 11.13 over SPRING, NAF, and FaStore, respectively.

In case of SRR125858 dataset, NAF remains the fastest among all algorithms. SCA-NGS is 62.72 and 54.84 percent faster than SPRING and FaStore, respectively. In terms of compression ratio, SCA-NGS achieves percent improvement gain of 0.34, 27.14, and 6.01 over SPRING, NAF, and FaStore, respectively. In terms of memory consumption, SPRING and FaStore consume more memory in comparison to other methods.

Percent improvement gain of SCA-NGS for dataset SRR61141 in terms of execution time over SPRING and FaStore is 86.2 and 85.84, respectively. Like other datasets, NAF is faster in all compared methods. In terms of compression ratio, percent improvement gain of SCA-NGS over the compared methods is 8.18, 20.34, and 12.9. SPRING and FaStore consume more memory both for compression and decompression.

The proposed algorithm performs better than all other algorithms in compression ratio for dataset SRR489793. NAF is faster in comparison to all compared methods. SCA-NGS achieves significant reduction in file size as compared to other methods. The results show that the proposed algorithm achieved percent improvement gain of 6516 and 56.91 over SPRING and FaStore respectively in terms of execution time. In terms of compression ratio percent improvement gain of SCA-NGS over SPRING, NAF, and FaStore is 8.17, 24.08, and 9.46, respectively.

In case of SRR935135 dataset, SCA-NGS achieved better results in terms of compression ratio and execution time except for NAF. NAF performs both compression and decompression in shorter execution time in comparison to other methods at the cost of compression ratio. Improvement gain attained by SCA-NGS in terms of compression ratio over SPRING, NAF, and FaStore is 4.74, 21.16, and 5.83 respectively. Improvement gain in terms of execution time over SPRING and FaStore is 77.7 and 67.62.

The proposed algorithm yields better compression ratio for the homo-sapiens dataset (SRR003177). Execution time is high as compared to NAF. There is a significant reduction in file size. The proposed algorithm achieved percent improvement gain in execution time of 86.2 and 85.89 over SPRING and FaStore, respectively. In terms of compression ratio, percent improvement gain of the proposed algorithm over the compared methods is 8.18, 20.34, and 12.9.

For the SRR40039 dataset, NAF yields the worst compression ratio with the fastest execution time and low memory consumption for compression. The proposed algorithm has slightly higher gain over SPRING. Percent improvement gain in terms of the compression ratio of SCA-NGS over SPRING, NAF, and FaStore is 1.19, 25, and 5.96 respectively. For the same dataset, the improvement gain in terms of compression time over SPRING and FaStore is 42.83 and 51.25.

Table 4 shows the comparative results of the proposed algorithm with Cryfa and GP²R algorithms. The results include encryption time in addition to parameters of the previous results. The results show that the proposed algorithm has much better compression ratio as compared to Cryfa and GP²R on all datasets. In most cases, the size of compressed data with the proposed algorithm is less than half of the size of the compressed data with Cryfa. Cryfa performs better in terms of execution time and memory. GP²R is faster and consumes less memory in comparison to SCA-NGS however compression ratio is worst in comparison to SCA-NGS.

Table 4.

Comparative experimental results of the proposed with Cryfa and GP²R on NGS datasets. CETime and DDTime refer to compression and encryption time and decompression and decryption times in seconds. CEMem and DEMem refer to compression and encryption and decompression decryption memory in MBs. CR refers to compression ratio.

Algorithms	CETime	CEMem	DDTime	DDMem	CR
SRR801793 (2818.11)
SCA-NGS	180	1148	58	1331	5.09
Cryfa	55.98	6	51	23	2.09
GP2R	75	352	55	302	3.56
ERR022075 (11253.16)
SCA-NGS	552	1131	305	1528	5.48
Cryfa	477	6	442	22	1.96
GP2R	512	371	431	317	3.51
SRR125858 (52172.64)
SCA-NGS	2437	1638	1531	2132	4.76
Cryfa	2047	6	2362	23	1.99
GP2R	2144	364	2016	342	3.46
SRR611141 (1799.86)
SCA-NGS	102	948	36	1142	4.03
Cryfa	41	6	37	21	1.94
GP2R	44	342	35	302	3.69
SRR489793 (13132.48)
SCA-NGS	876	1536	490	1562	4.65
Cryfa	635	7	716	21	1.91
GP2R	691	345	635	331	3.51
SRR935126 (10039.24)
SCA-NGS	412	1126	193	1433	5.33
Cryfa	322	6	321	21	1.96
GP2R	354	332	321	302	3.64
SRR003177 (1672.78)
SCA-NGS	68	1638	26	1532	5.12
Cryfa	50	8	53	28	1.85
GP2R	62	352	45	321	3.58
SRR400039 (65723.77)
SCA-NGS	3231	1556	2580	1752	4.14
Cryfa	2387	7	2413	21	2.0
GP2R	2561	348	2525	341	3.67

Note: Bold entries show best results for each parameter.

Conclusion

Data compression is extensively studied topic in bioinformatics. Many tools have been developed to compress biological data. Proper measures are required to maintain its confidentiality, integrity, and authenticity. Previously some encryption techniques were applied but they proved to be futile enough to gain good compression advantage. The compression ratio was much lower than other state-of-the-art algorithms. Encryption adversely impacts the execution time, memory acquisition. We have proposed a secure method for encryption and compression of NGS data. Our compression ratio with encryption is comparatively better as compared to previous algorithms. However, there could still be improvements over encryption time. Results vary over different architectures and processors. The nature of the dataset also impacts the overall compression gain. The security of data has been applied through crossover and mutation operations of the genetic algorithm. The compression of data has been applied by the grouping of data to different ASCII values and later transforming them into binary values. A general-purpose compression library is used to compress different components.

Author biographies

Muhammad Sardaraz received his master’s degree in computer science from Foundation University Islamabad. He completed his PhD in Computer Science in 2016 from Iqra University Islamabad, Pakistan. He worked as a lecturer in the Department of Computer Science, University of Wah, Wah Cantt, Pakistan. Presently Dr. Sardaraz is working as an assistant professor in the Department of Computer Science COMSATS University Islamabad, Attock campus, Pakistan. His research interests are cloud computing, cluster and grid computing, and bioinformatics.

Muhammad Tahir completed PhD (Computer Science) from the Department of Computing and Technology, Iqra University, Islamabad, Pakistan in 2016. He worked as a lecturer in the Department of Computer Science, University of Wah, Wah Cantt, Pakistan. He is currently working as an assistant professor in the Department of Computer Science COMSATS University Islamabad, Attock campus, Pakistan. His research interests are in parallel and distributed computing, Hadoop Mapreduce framework, IoT, bioinformatics.