Quantifying biallelic expression from single cell primary fibroblasts

We analyzed the published scRNA-Seq data from female primary human fibroblasts [28]. Before analysis, we took care of an experimental pitfall relevant to many single-cell studies. The pitfall concerns cell doubles in which more than one cell is sequenced per one library. In such a scenario, different active X-chromosomes (Xa) from two different cells that are included in the sample will produce a biallelic signal along the entire X-chromosome. While the fraction of doublets is expected to be small, it may lead to wrong interpretation [29]. Therefore, before analyzing the data from the fibroblasts, we revisited all 104 fibroblasts and tested their biallelic ratio with respect to ChrX (see Methods). Three cells showed an exceptionally high degree of biallelic expression that might indicate a mixture of two parental X-chromosomes (Additional file 1: Text and Figure S3). We removed all three suspicious cells from all the analyses.

Next, for every single cell, we counted the number of reads that were uniquely mapped to hSNP alleles. The allelic ratio (AR) for each iSNP is defined as the fraction of the reads mapped to the alternative allele (Alt) out of the total reads (see Methods, Additional file 2: Table S1). Figure 2a-c summarizes the AR of ChrX, Chr17, and the entire autosomal chromosomes according to the primary fibroblasts collection (101 out of 104 cells). In addition, Fig. 2d shows the distribution of the AR of an annotated set of imprinted genes from skin tissues (according to [27]). As previously reported [30, 31], a bias in mapping towards the reference genome (AR=0) is evident (Fig. 2a-d). Additionally, a substantial fraction of monoallelic expression was observed for all tested sets (Fig. ​(Fig.2a-d).2a-d). This dominant appearance of monoallelic expression in single cells is caused by a combination of both under-sampling of transcripts, and a phenomenon that is known as “transcriptional bursting” [28, 32–34].

Open in a separate window

Fig. 2

The distribution of the allelic ratio (AR) for each SNP as a fraction of the assignments for Alternative (Alt) out of Alt and Reference (Ref) alleles. The X-axis in the top histograms ranges from 0 to 1.0, where 0 indicates that all assignments are associated with the Ref allele and 1 indicates all assignments for the Alt allele. As the majority of the iSNPs are assigned with AR values of 0 or 1, each analysis is shown by two histograms. The lower histogram focuses on non-monoallelic iSNPs and covers all AR values excluding the AR=0 and AR=1. The percentage of iSNPs that are included in the lower histograms are shown. The distributions of the AR are shown for Chr17 (a), Autosomal chromosomes (b), ChrX (c) and imprinted genes (d). For source data, see Additional file 3: Table S2

We focused only on iSNPs that show a non-monoallelic signature (i.e., excluding AR=0 and AR=1). We observed a marked difference in the AR distribution of ChrX and imprinted genes relative to Chr17 and all autosomal chromosomes (compare Fig. ​Fig.2a2a to b and Fig. ​Fig.2c2c to d). Accordingly, several observations from the results shown in Fig. ​Fig.22 can be drawn: (i) Chr17 and all autosomes share a similar AR profile. (ii) A clear tendency towards balanced expression (AR=0.5) is apparent for any autosomal chromosomes (Fig. ​(Fig.2a-b),2a-b), but not ChrX or imprinted genes (Fig. ​(Fig.2c-d).2c-d). (iii) The fraction of non-monoallelic expression in autosomal chromosomes is higher (~18%) relative to ChrX (~9%). (iv) The fraction of non-monoallelic expression in imprinted genes shows an intermediate level (13%). Such an intermediate level is probably a reflection of the inherent inconsistency in the identity of imprinted genes [35, 36]. Additional file 3: Table S2 lists the supportive iSNPs for all the analyzed chromosomes in fibroblasts including the imprinted gene set.

Identifying escapees in single cell primary fibroblasts

In the dataset of the primary fibroblasts, there are 232 and 485 genes that are supported by iSNPs evidence for ChrX and Chr17, respectively. As these cells lack information of genome phasing (Fig. ​(Fig.1b),1b), information on escaping from X-chromosome is limited to the set of biallelic iSNPs (see Fig. ​Fig.1b).1b). We aggregated the iSNPs according to their corresponding genes (Fig. ​(Fig.1b).1b). The aggregation is performed across different single cells and across multiple iSNPs within a specific cell-gene pair. A gene will be labeled escapee candidate when it is associated with multiple biallelic iSNPs. Altogether we identified 24 such genes (Table 1) which account for 10.3% of all expressed genes in ChrX. As expected, the fraction of genes on Chr17 showing biallelic expression is substantially higher (49.3%, Additional file 4: Table S3).

Table ​Table11 lists the escapee candidates at varying degrees of support. For example, ZFX (Zinc finger X-chromosomal protein) and SMC1A (Structural maintenance of chromosomes protein 1A) genes are strongly supported with 103 and 19 biallelic iSNPs, respectively. A further increase in the reliability of identifying escapees is based on having at least 2 independent cells that contributed information on biallelic expression. We show that 21 out of 24 genes met this strict criterion (Table ​(Table1).1). Notably, among the identified escapees we detected only three PAR genes (SLC25A6, CD99, and DHRSX, Table ​Table1).1). The assignment of these genes as escapees agrees with the expected PAR expression. From the number of biallelic PAR genes out of the expressed PAR genes, we estimated the false negative discovery rate for escapees to be as high as 70% (i.e., missed 7 of 10 expressed PAR genes). Additional file 4: Table S3 shows the support for Table ​Table11.

Quantifying allelic expression from clonal phased lymphoblasts

A major limitation in the protocol described above concerns the lack of parental haplotype phasing. Under this setting, iSNPs cannot be assigned to Xa or Xi. Consequently, the random choice of Xi which characterizes primary cells limits the discovery rate for escapees. We expanded the analysis of scRNA-Seq to female-originated lymphoblasts from the clonal cell-line GM12878 [37]. See Fig. ​Fig.1c1c and Additional file 1: Figure S4.

Figure 3a shows the expression profile for clonal lymphoblast single cells (n=25) (Additional file 2: Table S1, Additional file 5: Table S4). In any single cell, the monoallelic expression reflects the combination of an under-sampling of transcripts and the phenomenon of transcriptional bursting [28, 32, 33, 38]. It is clear that the maternal expression from the Xa dominates (Fig. 3a, top). An observation that agrees with the reported maternal Xa origin of cell-line GM12878 [37]. In most of the analyzed cells, a small but substantial fraction of the measured total expression is from the paternal, Xi chromosome (Fig. ​(Fig.3a,3a, top). In contrast, Chr17 and the autosomal chromosomes show an equal expression from both alleles (Fig. ​(Fig.3a,3a, middle and bottom panels).

Open in a separate window

Fig. 3

Quantifying the iSNPs’ labels from 25 single cell lymphoblasts. a Each single cell is partitioned according to its tagged allelic iSNPs on ChrX, Chr17, and all autosomal chromosomes. The iSNPs are associated with maternal (pink), paternal (light blue) and balanced expression (gray). The cells are ordered from left to right according to their iSNP contributions (Additional file 1: Figure S4). On the right, the summary statistics of the Pool100 is shown. b A summary of the partition of iSNPs tags for all 25 single cells on ChrX, Chr17 and autosomal chromosomes. c A summary of the partition of iSNPs tags for Pool100. Blue and pink colors are associated with the paternal and maternal alleles, respectively. The striped pattern indicates biallelic iSNPs leaning towards paternal (blue) or maternal (pink) alleles. For single cells, the data is based on 375 iSNPs for ChrX, 808 iSNPs for Chr17 and 20,212 iSNPs for autosomal chromosomes. The data of Pool100 is based on 211 iSNPs for ChrX, 216 for Chr17 and 5360 iSNPs for autosomal chromosomes. For the source data see Additional file 5: Table S4

From the figure, it is evident that the phenomenon of transcriptional burst affects all chromosomes including ChrX. For assessing the impact of this phenomenon on identifying genes as escapees, we compared single cells with respect a pool of cells (Pool100, Fig. 3a, right bar). While the majority of the iSNPs from Chr17 display biallelic profiles, ChrX remains dominated by a maternal monoallelic expression.

Figure ​Figure3b3b is an aggregated view of ChrX, Chr17 and the autosomal chromosomes. The data are based on 375, 808 and 20,212 expressed hSNPs, respectively. Figure ​Figure3b3b (middle and right panels) shows an equal partition of the parental alleles from Chr17 and autosomal chromosomes (Fig. ​(Fig.3b,3b, top). Performing the same analysis on data collected from Pool100 (Fig. ​(Fig.3c)3c) shows that the partition of the parental alleles remains practically unchanged (compare the fraction occupied by pink and blue colors, Fig. 3b-c). Additionally, we observed a shift from a monoallelic (Fig. 3b-c, filled color) to a biallelic expression (Fig. ​(Fig.3b-c,3b-c, stripped color). The fraction of the biallelic expression for Ch17 increased from 19% in single cells to 80% in Pool100, and for autosomal chromosomes from 18 to 79% (Fig. ​(Fig.3c,3c, middle and right panels). The results from Pool100 indicate that the monoallelic expression observed in single cells is practically abolished by averaging the signal.

The results from ChrX (Fig. ​(Fig.3b3b (left) are fundamentally different relative to Chr17 or the autosomal chromosomes (Fig. ​(Fig.3b-c).3b-c). The most notable difference is that only 21% of the expressed iSNPs are associated with the paternal Xi allele in ChrX (Fig. ​(Fig.3b,3b, top). Furthermore, in analyzing Pool100, the fraction of biallelic expression remains bounded (a shift from 9% in single cells to 34% in Pool100). The observed pattern of ChrX from Pool100 (Fig. ​(Fig.3c,3c, left) is best explained by an averaging of the stochastic monoallelic signal (at the same degree as the other chromosomes) while maintaining a strong signal of the Xa monoallelic expression. See Additional file 5: Table S4 for lymphoblasts allelic ratio of all tested chromosomes and the Pool100.

Identifying escapees from single cell lymphoblasts

Figure 4a is a gene-centric view that shows the iSNP allelic partition from lymphoblasts (colored according to their origin as maternal, paternal or mixed expression, see Methods). Only the subset of genes that are supported by multiple iSNPs is listed according to their ordered along the chromosomes. Altogether we report on 93 annotated genes on ChrX (Fig. 4a, 30 escapees and 63 inactivated genes). Note that the X-inactivated genes account for genes which are expressed primarily from the maternal Xa. A cluster of genes with a paternal expression at the tip of ChrX p-arm represents the expected biallelic expression from the PAR genes (Fig. ​(Fig.4a).4a). Additional evidence for paternal expression is localized to the XIC with genes such as XIST, JPX, and FTX. While most of the escapees are supported by a limited number of iSNPs, a few of them such as ZFX, CD99, and SLC25A6 are supported by a relatively large number of supporting iSNPs (48, 38 and 34, respectively).

Open in a separate window

Fig. 4

A gene-centric partition of alleles from lymphoblast cells. a For each gene on ChrX, the iSNPs parental partition is shown along with the number of iSNPs. For clarity, only genes that are supported by >=2 iSNPS are listed. A total of 93 genes in ChrX are listed by their order on the chromosome. The color code is according to the iSNP labels as paternal, maternal and balanced expression. For the source data, see Additional file 5: Table S4. b-c Correlation between the expression levels from the paternal and maternal alleles. The scatter plots show the expression levels of genes by the number of reads associated with maternal (x-axis) and paternal (y-axis) alleles. The number of analyzed genes for each scatter plot is indicated (on the x-axis, in parenthesis). Data shown are from Chr17 (b) and ChrX (c) based on single cells and Pool100. Note that the number of reads for the Pool100 data is 10-fold less with respect to the cumulative data extracted from single cells. For the source data, see Additional file 4: Table S3

An alternative method for assessing the extent of the phenomenon of X-inactivation is by quantifying the evidence directly from the sum of all sequenced reads (abbreviated as the read-based protocol). Figure 4b-c compare read counts from Chr17 (Figs. 4b) and ChrX (Fig. ​(Fig.4c)4c) by the paternal versus maternal origin. We compare the expression data from the single cells and the Pool100. The linear regression for the gene expression from Chr17 shows a high correlation fit-line (r²=0.823, Fig. ​Fig.4b).4b). As expected, the correlation is stronger in the data originated from the Pool100 (r²=0.946, Fig. ​Fig.4b).4b). We concluded that despite the monoallelic expression due to the transcriptional bursting phenomenon, balanced allelic expression of all genes is strongly supported. For ChrX however, the resulting linear regression of the single cells is poor (r²=0.238, Fig. ​Fig.4c),4c), and was not improved by the data from the Pool100 (r²=0.222, Fig. ​Fig.4d).4d). Inspecting the expression data for ChrX shows that the regression lines actually lean toward the maternal Xa expression (x-axis). The expression data are consistent with two distinct regression lines for ChrX. One that matches the inactivated genes (parallel to the x-axis), and the other matches a trustfully biallelic expression.

Applying the conservative iSNP-based protocol leads to the identification of 30 genes as escapee candidates that are also supported by the read-based protocol (Table 2). The read-based (i.e., labeling a gene as escapee by having a minimal number of paternal reads, see Methods) protocol expanded the escapee candidate list to include overall 49 genes (Additional file 4: Table S3).

Testing the parental origin of alleles along a gene in the same cell is a strict test for the reliability of the iSNPs. This test is only valid for genes with multiple hSNPs. Such genes that are supported with two or more expressing hSNPs account for 44% of the genes. We consider a gene to be consistent if the expression along the gene in a specific cell is not monoallelic to both alleles. Altogether, we identified 3 inconsistent genes - TEX11, FTX, and ZCCHC16. For another 6 genes, the inconsistency is only partial as there are other observations of biallelic expression. The estimate from full inconsistency (3 out of the 29 genes that were eligible for this test) suggests that an upper limit for a faulty interpretation of 10%. Additionally, the iSNP-based protocol identified 9 out the 11 expressed PAR genes. Thus, we extrapolate the escapee detection rate to be 82%. Interestingly, analyzing Chr17, under the assumption that there is no systematic allelic bias [35], showed that 7.3 and 9.6% of the genes were associated with maternal and paternal monoallelic expression, respectively. These results provide an upper limit of 17.9% to the likelihood of false gene labeling in Chr17 and can be used for estimating the limitation of the method.

Demanding paternal evidence from at least two cells reduces the number of escapees from 49 to 18 (including 5 PAR genes). Many single evidence genes resulting from a specific outlier cell (SRR764803) that contributed to 17 of the 49 reported genes (Table ​(Table2,2, Additional file 5: Table S4). This cell exhibits the highest expression levels on both ChrX and Chr17 (Additional file 1: Text and Figure S4). Table ​Table22 lists the discovered escapee candidates along with their supporting evidence (Table ​(Table22).

Comparison of the identified escapees to current knowledge

We tested the correspondence between the identified escapees from our study and a literature-based catalog [13]. This unified catalog was compiled from the integration of four independent studies that covers 1144 genes from ChrX. The genes in this catalog are manually partitioned into nine defined categories (see Methods). The largest one accounts for genes that lack information (45%) [13]. About 15% of the genes (168/1144) are considered ‘escapee-associated’ (See Methods). We consider the compiled set of ‘escapee-associated’ genes as a gold standard to test escapee’s discovery rate in our study (total of 124 genes, excluded PAR genes, collectively called Balaton-Esc).

We applied a hypergeometric statistical test (see Methods) to assess the overlap of the different escapee gene lists (Fig. 5). Figure 5a shows the number of identified escapees from the fibroblasts and lymphoblasts (excluding PAR genes). Note that only genes that are included in the Balaton-Esc benchmark are included in this analysis (Fig. ​(Fig.5).5). Figure ​Figure5b5b shows the statistical significance of the overlap between the gene lists from Fig. ​Fig.5a5a and the Balaton-Esc [13]. As can be seen, there is a significant overlap between the escapees from lymphoblasts (Table ​(Table22 and Additional file 4: Table S3) and Balaton Esc list (Fig. ​(Fig.5b,5b, p-value=7.43E-8). Applying the same test for the primary fibroblasts (Table ​(Table1)1) resulted in a lower significance (p-value=4.07E-2).

Open in a separate window

Fig. 5

Identified escapees and statistical significance of the overlap with literature-based catalog compiled by Balaton et al. [13]. a The numbers of escapees identified by each of the analyses. The numbers include only genes that were present in Balaton et al. [13] and exclude PAR genes. b Statistical analysis based on the hypergeometric distribution measuring the overlap between the literature-based list as presented by Balaton et al. [13] and the escapee assigned in this study (as in (a)). Y-axis is the -log10(x) of the calculated p-value. c Venn diagram of the 4 sets of escapees according to the analyzed fibroblasts and lymphoblasts, the Balaton-Esc collection [13], and the Tukiainen-Esc [16]. Including PAR. For details see text. Source data in Additional file 6: Table S5

Figure ​Figure5c5c depicts the overlap genes between the escapees identified in our study and the Balaton-Esc catalog (168 genes including PAR genes) [24]. We also included a complementary resource based on 940 transcriptomes from scRNA-Seq (25 escapee genes, Tukiainen-Esc) [16]. The Venn diagram shows that each of the above studies contributes to the current knowledge on escapees. Escapees from the two external resources overlap by 18 out of 25 reported genes (72%). As shown in Fig. ​Fig.5c,5c, 62% of the escapees reported from the lymphoblasts overlap with the external escapee lists, while the fibroblasts are supported by only 38% overlap. Notably, most of our discovered candidate escapee genes from fibroblasts (62%) have no correspondence with the other tested lists (Additional file 6: Table S5).

LncRNAs extend the list of escapee candidates

We extended our analysis to long non-coding RNAs (lncRNAs) using the same escapee criteria as used from coding genes (Fig. 6). Altogether we identified 15 lncRNAs as escapee candidates, among them only a few were previously studied. The location of the lncRNAs and coding escapes along ChrX is shown (Fig. ​(Fig.6a).6a). We tested the positions of escapees along the ChrX relative to all ChrX genes. While the positional distribution for lncRNA escapees is similar (Kolmogorov–Smirnov test, p-value=0.57), it is different for coding escapees (Kolmogorov–Smirnov test, p-value=0.004, Fig. ​Fig.66a).

Open in a separate window

Fig. 6

LncRNAs assigned as escapees, and the groups of escapees according to their confidence level. a The dispersal of escapees along ChrX. Escapees belong to the lncRNAs and to the coding genes are indicated above and below the schematics of ChrX, respectively. b A table listing the 15 lncRNA escapee candidates. The orange color indicated escapee and the gray inactivated. An assignment that is based on a single iSNP is labeled with light orange. The white indicates lack of report or no expression. XIC, X inactivation center region. c Partition of the 75 genes that were mentioned as escapee candidates in this study. The categories are labeled ‘confirmed’, ‘approved’, ‘probable’ and ‘possible (see text). The two external resources that are used to define the groups are from the literature [13] and from single-cell by [16]. For a group of ‘probable’ escapee, we used escapee’s annotation from [10] as evidence. Note that 11 of the 15 listed lncRNA genes are included in the ‘possible’ escapee set. d A summary of the evidence-based groups for 75 genes. Genes are sorted according to the 4 evidence groups (as in c) The escapees are colored orange. Light orange indicates escapees that are supported by a single evidence from one cell only in fibroblasts. Inactivated genes are colored gray. White color indicates no expression or lack of report. The cell evidence is color coded showing support by one (light blue) or multiple cells (dark blue). Cases where evidence are based on Pool100 only are also marked light blue. PAR genes and lncRNAs are marked. Source data is in Additional file 6: Table S5

Figure 6b lists all 15 identified lncRNA escapee genes, among them, are ncRNA genes from the XIC that coordinate the activation and maintenance of X-inactivation. Many of the lncRNAs are localized at transcriptionally active segments (e.g., within the PAR or the XIC), while others are localized in non-conserved regions which are enriched with long and short ncRNAs. For additional lncRNAs, including inactivated genes see Additional file 7: Table S6.

Datasets

A dataset for scRNA-Seq of fibroblast UCF1014 was downloaded from the European Genome-phenome Archive (https://www.ebi.ac.uk/ega/home) using accession number EGAD00001001083. The data was produced by Borel et al. [28]. DNA-seq of UCF1014 was also downloaded from EGAD00001001084. The quality of the NGS data from scRNA-Seq in term of sequencing errors, PCR biases, and mapping errors was thoroughly discussed [28].

A dataset for scRNA-Seq of GM12878 cell-line from female lymphoblast NA12878 along with Pool100 was downloaded from the Gene Expression Omnibus (GEO) database (www.ncbi.nlm.nih.gov/geo) using the accession number GSE44618 [37]. The NA12878 phased genome was downloaded from http://sv.gersteinlab.org/NA12878_diploid/NA12878_diploid_2012_dec16/NA12878_diploid_genome_2012_dec16.zip. The SNP locations were downloaded from: http://sv.gersteinlab.org/NA12878_diploid/NA12878_diploid_2012_dec16/CEUTrio.HiSeq.WGS.b37.bestPractices.phased.hg19.vcf.gz.

For additional information and detailed protocols, see Additional file 1: Text.

RNA-Seq alignment

RNA-Seq from single cells (scRNA-Seq) and pooled cells were aligned to a relevant reference genome. Datasets of scRNA-Seq from fibroblasts were aligned against UCSC hg19 reference genome after removal of the Y chromosome. The lymphoid scRNA-Seq and RNA-seq are aligned against the Paternal and Maternal haplotypes of the phased diploid NA12878 genome at the hg19 format. Reads aligned to hSNPs were counted such that each of the reads is resigned to one of the two alleles. Each hSNP, expressed in every single cell with a minimal number of 7 aligned reads was considered an informative SNP (iSNP). For details see Additional file 1: Text. Quantitative values of the number of iSNPs for each cell of the 104 primary fibroblasts are shown in Additional file 1: Figure S2 and Additional file 2: Table S1. Quantitative values of the number of iSNPs for each cell of 25 single cell clonal lymphoblasts are in Additional file 1: Figure S4 and Additional file 2: Table S1.

Quantification and annotation of alleles by expression

For the primary fibroblasts without phased haplotype data, the Allelic Ratio (AR) of each iSNP is determined as follows:

Where c indicates a cell and i indicates an iSNP. #Alt and #Ref refer to the number of reads aligned to the alternative and the reference allele, respectively.

For the lymphoblasts that are associated with a phased genome, the Allelic Ratio (AR) of each iSNP is determined as follows:

Where c indicates a cell and i indicates an iSNP. #Pat and #Mat refer to the number of reads aligned to the paternal and maternal allele, respectively.

An iSNP with AR of <=0.1 or AR >0.9 is considered monoallelic and an iSNP associated with 0.1<AR≤0.9 is considered biallelic.

For the fibroblasts data, tagging a specific iSNP as escapee relies on annotating multiple iSNPs of a gene as biallelic. For the clonal lymphoblasts data, the paternal ChrX is associated with Xi. Therefore, iSNPs with paternal or biallelic AR (expression from the paternal ChrX) are indicative of an escapee gene. An additional protocol for identifying escapees from the clonal single cell lymphoblasts is based on accumulating reads across cells with an escapee criterion of a minimal number of 7 paternal reads per gene. Lymphoblast Pool100 cells were quantified such that in each gene, the sum of hSNP reads with a paternal expression ratio of AR >0.1 indicates an escapee gene. For further details, the quantitative values of the resources, and the NGS tools that were used see Additional file 1: Text (Sections 1–1.4).

Cell outliers according to extreme biallelic expression

A possible cause of faulty interpretation for biallelic expression of a gene can originate from a single cell experiment that accidentally included two (or more) cells instead of one cell. Therefore, if for a specific cell, the proportion of biallelic expression shows an extreme value, the data associated from that cell could potentially be explained by having a mixture of two cells expressing two different Xi (coined a doublet). We have calculated for each cell the biallelic ratio as biallelic ratio=biallelic iSNPs/total iSNPs. Following this investigation, we have excluded three single cells which had a biallelic ratio of over 3 standard deviations from the average of the entire cells. After filtering these cells outliers, 101 single cells remained for further analyses.

As the lymphoblasts were handpicked. No such procedure was needed.

A comparison to annotated ChrX gene catalog

There are 1144 known genes in ChrX that were annotated and compiled into a comprehensive catalog [13]. All genes were annotated and partitioned into 9 categories: (i) PAR, (ii) escapee, (iii) mostly-escapee, (iv) variable-escapee, (v) mostly-variable-escapee, (vi) discordant, (vii) inactivated genes, (viii) mostly-inactivated, and (ix) genes having no data. The set of criteria for assigning a gene to any of these categories was described in [13]. Annotation for the 9 categories is based on a careful analysis according to major publications combining numerous measurements for genes expression characteristics of ChrX [13]. The reports used for compiling the catalog were from four major resources extracted from [13, 17, 20, 46].

From the genes on ChrX, an “escapee associated” list was compiled from the 6 first categories (i-vi) associated with escapees, including PAR and discordant genes. This unified set corresponds to 15% of the genes, the rest are divided between inactivated-related categories (40%) and genes that are are annotated as “no data” (45%).

Statistical analysis

Hypergeometric probability between our results and the external annotated catalog was calculated by comparing the correspondence of any two lists of escapees. We used standard notations of N, k, n and x: N symbolizes all genes on ChrX from [13] with a label other than “No data”; k is the number of escapees by [13] which are associated with any escapee annotations (i.e. the escapee-associated); n is the number of escapees we identified by any of the settings from our protocols; x the number of genes in our list that match the literature-based escapee list in k. P(x) is the probability that an n-trial will result in a value that is > x. See Additional file 1: Text (Section 1.5).

We thank Nati Linial for useful discussions and critical comments. We thank Liran Carmel, Shahar Shohat and Michal Linial’s lab members for critical reading of the manuscript. We thank Yuval Nevo and the CSE system for technical assistance. KWK is a recipient of the Ariane de Rothschild Scholarship. This study used best practice protocols that are compiled in EU Elixir collection.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.