Thesis: Analysis of Embryo Genetic Scores

This thesis was presented in the Faculty of Exact and Natural Sciences of the University of Buenos Aires (UBA) in 2022. It’s a Bioinformatics and Biostatistics analysis of several kinds of genetic scores associated to disease, mainly utilised to diagnose embryos during in vitro fertilization.

Some results and figures are showcased here.

Theoretical description of genetic data under different disorders

We described the expected distribution of a genetic data statistic called BAF under different disorders and levels of mosaicism, which means different percentages of mutated cells in a sample.

Theoretical distribution of a statistic called BAF (B allele frequency) under different levels of mosaicism $m$, in monosomy and in trisomy.

We described the expected behavior of a single measure $\theta$ that helped designing statistical tests to easily detect these disorders.

Simulation of genetic data with different degrees of mosaicism

We simulated genetic data for different levels mosaicism, based on the theoretical framework developed for the thesis.

Simulated BAF ("B" allele frequency) for different levels of mosaicism $m$ in two different cases: monosomy (top) and trisomy (bottom).

Performance and empirical distribution of a new statistic

We developed and described a new statistic to detect aneuploidies in embryos, leveraging on a robust measure of dispersion called median absolute deviation (MAD).

Performance of a new statistic $T_3$ for the detection of genetic disorders under several scenarios (top). Empirical distribution of this statistic under the null hypothesis of a healthy embryo, to decide the threshold of detection (bottom).

Performance of testing with overlapping-windows of data

We described a problem with non-overlapping windows of data when searching for genetic disorders that span few megabases in a chromosome (that means, mutations that span a relatively small region).

In orange, a small region of a chromosome affected by a disorder called monosomy. The test statistic $T_1$ is maximized with overlapping windows (blue), favoring detection.

Estimator performance evaluation

We evaluated the performance of a new estimator of the level of mosaicism with simulated embryo samples.

Performance of the estimator $\hat{m}$ under different values of $m$ in two settings, trisomy and monosomy.

Complex patterns of DNA contamination

We analyzed the BAF statistic under different levels of DNA mixture (or “contamination”) between mother and embryo.

Patterns of BAF under different levels of mother-embryo DNA mixture. In colors the possible genotype combinations.

Genomic panel target distribution

We analyzed the distribution of “probes” or targets in a genomic panel, with emphasis in the distribution of the size of the gaps with no data left by the panel.

Schematic design of a panel of genomic targets (top). Distribution of the size of gaps left by the panel design (bottom).

Anticorrelation between Polygenic Risk Scores

We analyzed the correlation of Polygenic Risk Scores (PRS), a kind of genetic score computed from hundreds or thousands of mutations in the genome. We found several pairs of diseases, like rheumatoid arthritis and multiple sclerosis, with negatively correlated scores. An example is given in the figure below.

Rheumatoid Arthritis and Multiple Sclerosis have anticorrelated scores. We still see the correlation when converting the PRS to absolute risk (RA). A Student's $t$ test of paired samples shows a significant increase in risk when selecting individuals of low PRS.

Variance trajectories of sub-scores

We analyzed the proportion of variance $\mathcal{V}_{(k)}$ captured by the genetic scores when keeping sub-selections of the $k$ strongest-effect mutations. These captured-variance trajectories help quantify the degree of polygenicity of a given disease. Low polygenicity means that few mutations determine most of the risk.

Highlighted phenotypes in blue have pronounced captured-variance trajectories, this means they have low polygenicity.

Graph of diseases with correlated genetic risk

We found a surprising “natural” grouping of autoimmune diseases based on the correlation between their polygenic scores.

Directed graph of polygenic score correlations. The diseases naturally form two groups (left vs. right), with correlations only found inter-group and never intra-group.

Simulation of embryos and performance of embryo-selection strategies

We simulated genetic data from embryos and analyzed the results of selecting embryos based on their genetic scores. A potential problem was described where embryos with low genetic score of one disease might have increased risk of a correlated disease.

The whole set of simulated embryos is shown in grey, with the embryos of parents in the first quintile of PRS in orange. Blue dots are the low-risk embryos as selected by different strategies. The subgroup of selected embryos based on one score (X axis) has an increased absolute risk of a different disease (Y axis) under some strategies.

Relation between increase in risk and number needed to harm

A knwon metric in epidemiology called Number Needed to Harm (NNH) is plotted against the increase in absolute risk (IRA) for several pairs of anticorrelated diseases. This illustrates that between 250 and 1250 couples applying a PRS-based embryo selection might be enough to produce unwanted new cases of some diseases.

Relation between increase in absolute risk (IRA) and number needed to harm (NNH) for several pairs of diseases found to be anticorrelated in the thesis.

Heatmap of number of datapoints per genomic region

The number of SNPs in each window of 5 Mb (megabases, a kind of genetic distance) are plotted, which can be thought of as the number of datapoints for a given test.