Revising a personal genome by comparing and combining data from two different sequencing platforms

Abstract

For the robust practice of genomic medicine, sequencing results must be compatible, regardless of the sequencing technologies and algorithms used. Presently, genome sequencing is still an imprecise science and is complicated by differences in the chemistry, coverage, alignment, and variant-calling algorithms. We identified ~3.33 million single nucleotide variants (SNVs) and ~3.62 million SNVs in the SJK genome using SOLiD and Illumina data, respectively. Approximately 3 million SNVs were concordant between the two platforms while 68,532 SNVs were discordant; 219,616 SNVs were SOLiD-specific and 516,080 SNVs were Illumina-specific (i.e., platform-specific). Concordant, discordant, and platform-specific SNVs were further analyzed and characterized. Overall, a large portion of heterozygous SNVs that were discordant with genotyping calls of single nucleotide polymorphism chips were highly confident. Approximately 70% of the platform-specific SNVs were located in regions containing repetitive sequences. Such platform-specificity may arise from differences between platforms, with regard to read length (36 bp and 72 bp vs. 50 bp), insert size (~100-300 bp vs. ~1-2 kb), sequencing chemistry (sequencing-by-synthesis using single nucleotides vs. ligation-based sequencing using oligomers), and sequencing quality. When data from the two platforms were merged for variant calling, the proportion of callable regions of the reference genome increased to 99.66%, which was 1.43% higher than the average callability of the two platforms, representing ~40 million bases. In this study, we compared the differences in sequencing results between two sequencing platforms. Approximately 90% of the SNVs were concordant between the two platforms, yet ~10% of the SNVs were either discordant or platform-specific, indicating that each platform had its own strengths and weaknesses. When data from the two platforms were merged, both the overall callability of the reference genome and the overall accuracy of the SNVs improved, demonstrating the likelihood that a re-sequenced genome can be revised using complementary data.

Figure 1. Concordance of SNVs identified by the two different sequencing platforms.

Figure 2. Cumulative frequency plot of sequencing depths in heterozygous calls. The sequencing depths of heterozygous calls in the SOLiD data are plotted.

The patterns of concordant SNVs that are either chip-concordant or chip-discordant are almost compatible, which explains why the majority of heterozygous concordant SNVs that are chip-concordant are highly confident calls. In contrast, the median of discordant SNVs that are chip-discordant is substantially lower than those of concordant SNVs, which explains why only 25% of them are highly confident calls.

Figure 3. Callable and non-callable regions.

(A) Using Illumina and SOLiD data, 98.3% and 98.15% of the reference genome are callable, respectively. Using the merged data, the callability increases to 99.66%, which is 1.43% higher than the average callability of two platforms, representing about 40 million bases. (B) The base composition of non-callable regions. In SOLiD data, the proportions of A, T, C, and G were almost even. In Illumina data, the proportions of A and T were higher than those of C and G.