A Sequence Obfuscation Method for Protecting Personal Genomic Privacy

Abstract

With the technological advances in recent decades, determining whole genome sequencing of a person has become feasible and affordable. As a result, large-scale individual genomic sequences are produced and collected for genetic medical diagnoses and cancer drug discovery, which, however, simultaneously poses serious challenges to the protection of personal genomic privacy. It is highly urgent to develop methods which make the personal genomic data both utilizable and confidential. Existing genomic privacy-protection methods are either time-consuming for encryption or with low accuracy of data recovery. To tackle these problems, this paper proposes a sequence similarity-based obfuscation method, namely IterMegaBLAST, for fast and reliable protection of personal genomic privacy. Specifically, given a randomly selected sequence from a dataset of genomic sequences, we first use MegaBLAST to find its most similar sequence from the dataset. These two aligned sequences form a cluster, for which an obfuscated sequence was generated via a DNA generalization lattice scheme. These procedures are iteratively performed until all of the sequences in the dataset are clustered and their obfuscated sequences are generated. Experimental results on benchmark datasets demonstrate that under the same degree of anonymity, IterMegaBLAST significantly outperforms existing state-of-the-art approaches in terms of both utility accuracy and time complexity.

Keywords: DNA generalization lattice; IterMegaBLAST; MegaBLAST; clustering; genomic privacy; machine learning; obfuscation methods; sequence similarity.

FIGURE 1

Current status of personal genomic data for utilization and privacy. (A) The whole genome sequencing (WGS) cost decreased significantly with the technological advances in recent decades. (B) The number of COVID-19 cases increased significantly in these 2 years and concurrently the number of personal genomic data would increase. (C) Large-scale projects have been launched for betterment of human healthcare while simultaneously posing serious challenges on protecting individual genomic privacy. “All of Us” (https://allofus.nih.gov/) was launched by US and “1 + Million Genomes” (https://digital-strategy.ec.europa.eu/en/policies/1-million-genomes) was initialized by the European Union. Blue circles represent good benefits of genomic data utilization or privacy well proteted, whereas the red circle represents the challenge of genomic data privacy breached.

FIGURE 2

The generalization hierarchy (Malin, 2005b) for sequence or nucleotide obfuscation. Note that lev is the level of corresponding nucleotides and the symbol “-” represents the gap.

FIGURE 3

Statistics of the two datasets used in this paper. (A) and (B): The density distribution of the sequence lengths for Dataset I (A) and Dataset II (B). (C,D): Distributions of the percentages of each nucleotide (i.e., A, C, G, and T) for Dataset I (C) and Dataset II (D). The numbers of sequences for Dataset I and Dataset II are 56 and 372, respectively.

FIGURE 4

The average distances of IterMegaBLAST varying with respect to the number of DNA sequences for (A) Dataset I and (B) Dataset II. The shorter the distance is, the less the information loss. DNALA is from (Malin, 2005b), while MWM, Hybrid and Online algorithms are from (Li et al., 2012). IterMegaBLAST is the method proposed in this paper.

FIGURE 5

An example of how IterMegaBLAST works. IterMegaBLAST consists of two major steps: sequence alignment and clustering, and sequence obfuscation. Given a query sequence, e.g., LN∣AF392171∣GI∣18029617, IterMegaBLAST first uses MegaBLAST to find its top homolog, e.g., LN∣AF392284∣GI∣18029730, which form a cluster. Then, IterMegaBLAST uses the sequence obfuscation method introduced in Section 2.4 to generate the generalized sequence (see the “Sequence info” box on the right panel) for this cluster. The distance is calculated and the related meta information is produced (see the “Meta info” box on the right panel). The red circles indicate the mismatched nucleotides between the query sequence and the homolog, and the blue circles represent the generalized nucleotides for the mismatches nucleotides.