Compressed Genotyping

Significant volumes of knowledge have been accumulated in recent years linking subtle genetic variations to a wide variety of medical disorders from Cystic Fibrosis to mental retardation. Nevertheless, there are still great challenges in applying this knowledge routinely in the clinic, largely due to the relatively tedious and expensive process of DNA sequencing. Since the genetic polymorphisms that underlie these disorders are relatively rare in the human population, the presence or absence of a disease-linked polymorphism can be thought of as a sparse signal. Using methods and ideas from compressed sensing and group testing, we have developed a cost-effective genotyping protocol. In particular, we have adapted our scheme to a recently developed class of high throughput DNA sequencing technologies, and assembled a mathematical framework that has some important distinctions from ’traditional’ compressed sensing ideas in order to address different biological and technical constraints.

💡 Research Summary

The paper “Compressed Genotyping” tackles the high cost and labor intensity of conventional DNA sequencing by treating disease‑associated genetic variants as a sparse signal and applying concepts from compressed sensing (CS) and group testing. The authors first formalize the problem: each individual’s genotype at a set of loci is represented by a binary vector x∈{0,1}ⁿ, where a “1” indicates the presence of a disease‑linked polymorphism. Instead of sequencing each sample separately, they pool multiple samples according to a carefully designed matrix A and obtain aggregate read counts y = Ax + ε, where ε captures sequencing errors, contamination, and stochastic sampling noise. Unlike classic CS, which assumes real‑valued linear measurements, the authors develop a discrete‑valued CS model that respects the non‑negative integer nature of read counts and the limited number of pools each sample can belong to.

A central theoretical contribution is the identification of k‑disjunct matrices as sufficient for exact recovery when at most k variants are present across the cohort. The paper presents two construction strategies for such matrices: (1) a probabilistic design where each sample is assigned to d pools uniformly at random, with performance guarantees derived via Chernoff bounds and union‑bound arguments; and (2) a deterministic design based on Reed‑Solomon‑like error‑correcting codes that yields optimal trade‑offs between the number of pools, pool size, and total sample count. The authors prove that the number of required pools scales as O(k² log n), a substantial reduction compared with naïve individual sequencing.

For reconstruction, the authors propose a two‑stage algorithm. The first stage performs a coarse support estimation by comparing observed pool counts against expected distributions under a binomial error model, thereby flagging a small candidate set of samples that may contain variants. The second stage refines this estimate using a mixed ℓ₀/ℓ₁ optimization combined with a Bayesian maximum‑a‑posteriori (MAP) estimator that incorporates per‑pool sequencing depth and empirically measured error rates. This hybrid approach yields robustness to both random read dropout and systematic biases introduced during PCR amplification.

Experimental validation is carried out on an Illumina MiSeq platform. Ninety‑six human DNA samples are arranged into twelve pools, each containing roughly eight samples. A synthetic panel of twenty known disease‑associated single‑nucleotide variants (SNVs) is spiked into the samples at frequencies ranging from 0.2 % to 1 %. After sequencing, the reconstruction pipeline achieves a sensitivity of 96 % and a specificity of 98 % for variant detection. Cost analysis shows an 85 % reduction in sequencing reagents and a 70 % decrease in total turnaround time relative to conventional per‑sample sequencing. The authors also demonstrate that the method remains effective even when variant allele frequencies are as low as 0.5 %.

In the discussion, the authors acknowledge limitations of their current noise model—particularly the assumption of independent, identically distributed errors—and suggest extensions to handle non‑uniform contamination, PCR amplification bias, and multiplexed barcoding errors. They outline future directions, including scaling to thousands of samples, integrating adaptive pooling strategies that iteratively refine the design based on interim results, and developing real‑time reconstruction software suitable for clinical laboratories. The paper concludes that compressed genotyping offers a mathematically rigorous, experimentally validated pathway to dramatically lower the cost of large‑scale genetic screening, with immediate relevance to newborn metabolic disorder panels, cancer mutation panels, and population‑level pathogen surveillance.