Small-sample estimation of the mutational support and the distribution of mutations in the SARS-Cov-2 genome

Vishal Rana, Eli Chien, Jianhao Peng, Olgica MilenkovićmedRxiv 2020 · DOI

Abstract

The problem of estimating unknown features of viral species using a limited collection of observations is of great relevance in computational biology. We consider one such particular problem, concerned with determining the mutational support and distribution of the SARS-Cov-2 viral genome and its open reading frames (ORFs). The mutational support refers to the unknown number of sites that is expected to be eventually mutated in the SARS-Cov-2 genome. It may be used to assess the virulence of the virus or guide primer selection for real-time RT-PCR tests during the early stages of an outbreak. Estimating the unknown distribution of mutations in the genome of different subpopulations while accounting for the unseen may aid in discovering adaptation mechanisms used by the virus to evade the immune system. To estimate the mutational support in the small-sample regime, we use GISAID sequencing data and new state-of-the-art polynomial estimation techniques based on weighted and regularized Chebyshev approximations. For distribution estimation, we adapt the well-known Good-Turing estimator. We also perform a differential analysis of mutations and their sites across different populations. Our analysis reveals several findings: First, the mutational supports exhibit significant differences in the ORF6 and ORF7a regions (older vs younger patients), ORF1b and ORF10 regions (females vs males) and as may be expected, in almost all ORFs (for Asia versus Europe and North America). Second, despite the fact that the N region of SARS-Cov-2 has a predicted 10% mutational support, almost all observed mutations fall outside of the two regions of paired primers recommended for testing by the CDC.

Introduction: The Small-Sample Challenge

During the early stages of a viral outbreak, researchers face a critical "small-sample" problem. With only a few thousand initial viral sequences available, the virus is destined to infect millions and mutate across tens of thousands of sites. Standard statistical tools, such as Maximum Likelihood Estimators, are often "greedy" because they only account for mutations already observed. They fail to account for the "unseen": genomic sites that have not yet mutated in the current sample but are statistically likely to do so as the pandemic progresses.

Accurately predicting this "mutational support" is vital for maintaining diagnostic integrity (ensuring PCR primers do not target rapidly mutating regions), tracking virulence, and understanding how the virus adapts across different host demographics.

Methodology: Predicting the Unseen

The researchers framed mutation prediction as a support estimation problem, drawing from information theory. The core of their approach is the Regularized Weighted Chebyshev (RWC) Estimator. While standard estimators use Chebyshev polynomials to approximate unseen distributions, they can be numerically unstable. The authors introduced a regularization term to balance bias and variance, solved efficiently via a Linear Program (LP). This allows the model to scale to the ~30,000 base pair SARS-CoV-2 genome.

Overview of the computational pipeline for estimating mutational support from small-sample sequencing data.

Additionally, the team adapted the classic Good-Turing estimator to determine the frequency distribution of mutations. This pipeline allows for a differential analysis of mutations across various subpopulations by processing raw sequencing data from the GISAID database.

Data and Analysis

The study utilized viral sequencing data from the GISAID EpiCoV database, starting with approximately 9,000 samples collected early in the pandemic (April 2020). To verify the accuracy of these early "small-sample" predictions, the authors later validated their findings against larger datasets containing over 115,000 samples from North America and Europe. All mutations were indexed against the "Patient 1" reference genome (Wuhan-Hu-1).

Key Findings and Genomic Insights

The analysis demonstrated that the "unseen" mutational capacity of a virus is quantifiable without needing to sequence every patient. The RWC estimator effectively identifies the statistical shadow of future mutations.

Structural map of the SARS-CoV-2 genome highlighting the various open reading frames (ORFs) analyzed for mutational support.

Demographic and Geographic Signatures

Mutations are not distributed uniformly. The study found that mutational support in the ORF1b region for young females was nearly double that of older males, suggesting different selective pressures based on the host's immune profile. Furthermore, Asian samples showed significantly higher mutational support early on, providing statistical evidence that the virus had been circulating in that region longer before detection.

Diagnostic Robustness

A significant finding concerned the N (nucleocapsid) region. Despite being identified as a mutational hotspot with a predicted 10% support, the specific genomic sites chosen by the CDC for PCR primers were located in rare "cold spots" within that gene. This explains why these diagnostic tests remained robust even as the virus evolved.

Conclusion and Future Directions

While the current model assumes samples are independent and identically distributed (I.I.D.)—ignoring the complex transmission chains of viral phylogeny—it provides a powerful tool for early-stage surveillance. Future work aims to integrate this framework into real-time genomic monitoring to identify "primer-breaking" mutations before they become widespread. By predicting where mutations can happen, rather than just where they have happened, public health officials can stay one step ahead of viral evolution.

Summary updated

Abstract

A bstract The problem of estimating unknown features of viral species using a limited collection of observations is of great relevance in computational biology. We consider one such particular problem, concerned with determining the mutational support and distribution of the SARS-Cov-2 viral genome and its open reading frames (ORFs). The mutational support refers to the unknown number of sites that is expected to be eventually mutated in the SARS-Cov-2 genome. It may be used to assess the virulence of the virus or guide primer selection for real-time RT-PCR tests during the early stages of an outbreak. Estimating the unknown distribution of mutations in the genome of different subpopulations while accounting for the unseen may aid in discovering adaptation mechanisms used by the virus to evade the immune system. To estimate the mutational support in the small-sample regime, we use GISAID sequencing data and new state-of-the-art polynomial estimation techniques based on weighted and regularized Chebyshev approximations. For distribution estimation, we adapt the well-known Good-Turing estimator. We also perform a differential analysis of mutations and their sites across different populations. Our analysis reveals several findings: First, the mutational supports exhibit significant differences in the ORF6 and ORF7a regions (older vs younger patients), ORF1b and ORF10 regions (females vs males) and as may be expected, in almost all ORFs (for Asia versus Europe and North America). Second, despite the fact that the N region of SARS-Cov-2 has a predicted 10% mutational support, almost all observed mutations fall outside of the two regions of paired primers recommended for testing by the CDC. Author Summary We introduce the new problem of small-sample estimation of the number of mutations and the distribution of mutations in viral and bacterial genomes, and in particular, in the SARS-Cov-2 genome. The approach is of interest due to the fact that it aims to predict which regions in the genome will mutate in the future and with what frequency, given only a very limited number of complete viral sequences. This setting is usually encountered during the early stages of an outbreak when it is critical to assess the potential of the virus to gain mutations advantageous for its spreading. The results may also be used to guide the selection of genomic (primer) regions that are not subject to mutational pressure and can consequently be used as identifiers in the process of testing for the disease. They can also highlight differences in the mutation rates and locations of the SARS-Cov-2 virus affecting diverse subpopulations and therefore potentially suggest the role of certain mutations in evading the immune system. Our approach uses a new class of estimation methods that may find other applications in bioinformatics.
BibTeX
@misc{rana2020small,
  author = {Vishal Rana and Eli Chien and Jianhao Peng and Olgica Milenković},
  title = {{Small-sample estimation of the mutational support and the distribution of mutations in the SARS-Cov-2 genome}},
  year = {2020},
  doi = {10.1101/2020.04.23.20076075},
  howpublished = {medRxiv},
}

Loading related content...

Figure