cell growth experiments or luciferase assays can then be performed to demonstrate the
biological consequences of the variant.
This type of filtering methodology has been extensively applied by projects such as the
Deciphering Developmental Disorders (DDD) study [21]. This Consortium applied
microarray and exome sequencing technologies to 1,133 trios (affected children and their
parents) and was able to increase by 10% the number of children that could be diagnosed,
as well as identifying 12 novel causative genes [22]. It may also be useful for the detection
of causal genetic variation for neurodevelopmental disorders [23]. However, while
tremendously useful in the cases where the three sequences are available, and where the
variant is present in the child and not the parents, this strategy is not that useful in those
cases where the causal variant may also be present in the parents or where it has a lower
penetrance.
Another definition of a “de novo” variant may be one that has never before been seen in
a population, which is identified through comparisons against population variation
databases such as gnomAD [15] and dbSNP [24]. Sometimes, researchers assume that a
rare variant, because it is rare (and perhaps because it falls in a biologically relevant gene)
then it must underlie their phenotype of interest. However, this is nearly always not true:
Depending on ancestry, estimates are that humans can carry up to 20,000 “singletons” (this
is, genetic variants only observed once in a dataset) [25] and can carry more than 50 genetic
variants that have been classified as disease-causing [15]. This point has been beautifully
illustrated by Goldstein and colleagues [26]: They analyzed sequencing data from a control
sample and reported finding genetic variants falling in highly conserved regions from
protein-coding genes, that have a low allelic frequency in population databases, that have a
strong predicted effect on protein function and in genes that can be connected to specific
phenotypes in disease databases. However, even if fulfilling all these criteria, these variants
clearly do not have a phenotype. They call the tendency of these kind of variants to be
assumed as causal as “the narrative potential,” which is unfortunately common in the
literature [27]. Therefore, in the next section we will summarize the aspects that need to be
taken into account in order to confidently assign a genetic variant as causative for a
phenotype.
10.6 Filtering Genetic Variants to Identify Those Associated to
Phenotypes
Given the huge number of genetic variants usually identified in NGS studies (12,000 in
exomes, ~5 million in genomes) [28], filtering and post-processing to pinpoint candidates
may be the most labor-intensive tasks out of the whole analysis pipeline. Depending on the
researcher’s biological question, they may need to tune these parameters to better answer it.
For example, if they are searching for rare variation in pedigrees that may predispose to a
disease, they may want to set quite permissive quality thresholds so as to not lose any
potential candidates, but making sure that any potential variants are confirmed through re134
P. Basurto-Lozada et al.
biological consequences of the variant.
This type of filtering methodology has been extensively applied by projects such as the
Deciphering Developmental Disorders (DDD) study [21]. This Consortium applied
microarray and exome sequencing technologies to 1,133 trios (affected children and their
parents) and was able to increase by 10% the number of children that could be diagnosed,
as well as identifying 12 novel causative genes [22]. It may also be useful for the detection
of causal genetic variation for neurodevelopmental disorders [23]. However, while
tremendously useful in the cases where the three sequences are available, and where the
variant is present in the child and not the parents, this strategy is not that useful in those
cases where the causal variant may also be present in the parents or where it has a lower
penetrance.
Another definition of a “de novo” variant may be one that has never before been seen in
a population, which is identified through comparisons against population variation
databases such as gnomAD [15] and dbSNP [24]. Sometimes, researchers assume that a
rare variant, because it is rare (and perhaps because it falls in a biologically relevant gene)
then it must underlie their phenotype of interest. However, this is nearly always not true:
Depending on ancestry, estimates are that humans can carry up to 20,000 “singletons” (this
is, genetic variants only observed once in a dataset) [25] and can carry more than 50 genetic
variants that have been classified as disease-causing [15]. This point has been beautifully
illustrated by Goldstein and colleagues [26]: They analyzed sequencing data from a control
sample and reported finding genetic variants falling in highly conserved regions from
protein-coding genes, that have a low allelic frequency in population databases, that have a
strong predicted effect on protein function and in genes that can be connected to specific
phenotypes in disease databases. However, even if fulfilling all these criteria, these variants
clearly do not have a phenotype. They call the tendency of these kind of variants to be
assumed as causal as “the narrative potential,” which is unfortunately common in the
literature [27]. Therefore, in the next section we will summarize the aspects that need to be
taken into account in order to confidently assign a genetic variant as causative for a
phenotype.
10.6 Filtering Genetic Variants to Identify Those Associated to
Phenotypes
Given the huge number of genetic variants usually identified in NGS studies (12,000 in
exomes, ~5 million in genomes) [28], filtering and post-processing to pinpoint candidates
may be the most labor-intensive tasks out of the whole analysis pipeline. Depending on the
researcher’s biological question, they may need to tune these parameters to better answer it.
For example, if they are searching for rare variation in pedigrees that may predispose to a
disease, they may want to set quite permissive quality thresholds so as to not lose any
potential candidates, but making sure that any potential variants are confirmed through re134
P. Basurto-Lozada et al.