OTAR3088/CellFinder-all · Datasets at Hugging Face

sentences stringlengths 0 2.55k	entities listlengths 0 23	file_id stringclasses 10 values	ner listlengths 1 1
GO categories are ordered from top to bottom in order of decreasingly significant bias toward enriched genes.(A) GO analysis of enriched genes in hESCs.(B) GO analysis of enriched genes in NPs.	[ { "end": 192, "label": "CellType", "start": 189 }, { "end": 151, "label": "CellType", "start": 146 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
To summarize, firstly immunohistochemical and RT-PCR evidence validated that the cells exhibited expected characteristics; secondly, stage-specific marker gene differences by RT-PCR were reflected accurately by gene-level estimates from the exon arrays; thirdly, the hESC-enriched genes were coherent with known genes th...	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Description of the Regression-Based Exon Array ProtocolConvinced that the signal estimates from the exon arrays reflected expected molecular and biological differences between hESCs and NPs, we sought to identify AS events.	[ { "end": 189, "label": "CellType", "start": 186 }, { "end": 181, "label": "CellType", "start": 176 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellType", "...
We compared Cyt-ES to hCNS-SCns to illustrate our approach.	[ { "end": 29, "label": "CellType", "start": 22 }, { "end": 18, "label": "CellLine", "start": 12 }, { "end": 31, "label": "CellType", "start": 22 } ]	17967047	[ { "tags": [ "O", "O", "B-CellLine", "O", "B-CellType", "O", "O", "O", "O", "O" ], "tokens": [ "We", "compared", "Cyt-ES", "to", "hCNS-SCns", "to", "illustrate", "our", "approach", "." ...
First we normalized the data and generated signal estimates with Robust Multichip Analysis (RMA) and estimated the probability that each probeset was detected above background (DABG) using publicly available Affymetrix Power Tools (APT).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
We analyzed probesets that (i) comprised three or more individual probes; (ii) were localized within the exons of our gene models with evidence from at least three sources (mRNA, EST, or full-length cDNA); and (iii) were detected above background in at least one of the cell lines.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
In total, 17,430 gene models were represented by probesets that satisfied these criteria.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "In", "total", ",", "17,430", "gene", "models", "were", "represen...
Next we asked whether the probeset expression within each gene model was positively correlated for any two cell lines.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Next", "we", "asked", "whether", ...
To do this we calculated the Pearson correlation coefficient between the vectors of median signal estimates across replicates in Cyt-ES versus hCNS-SCns.	[ { "end": 150, "label": "CellType", "start": 143 }, { "end": 152, "label": "CellType", "start": 143 }, { "end": 135, "label": "CellLine", "start": 129 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "B-CellType", "O" ], "tokens": [ "...
The vast majority of genes (>80%) was found to have probeset-level Pearson correlation coefficients of greater than 0.8 (Figure 3A).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ...
Next we randomly permuted the association between the median signal estimates and the probesets for each gene in hESCs (or hCNS-SCns) and observed that the distribution of Pearson correlation coefficients for the permuted sets was centered at zero, as expected (Figure 3A).	[ { "end": 130, "label": "CellType", "start": 123 }, { "end": 132, "label": "CellType", "start": 123 }, { "end": 118, "label": "CellType", "start": 113 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "B-CellType", "O", "O", "O", "...
This indicated that the signal estimates for probesets between hESCs and hCNS-SCns were highly correlated and suggested that a scatter plot of probeset signal estimates between hESCs and hCNS-SCns would reveal a linear relationship for the majority of genes.	[ { "end": 182, "label": "CellType", "start": 177 }, { "end": 82, "label": "CellType", "start": 73 }, { "end": 196, "label": "CellType", "start": 187 }, { "end": 68, "label": "CellType", "start": 63 }, { "end": 194, "label": "CellType", "star...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "...
We hypothesized that a linear regression to determine if some probesets behaved unexpectedly in one cell type compared to the other might be a reasonable approach to identify AS exons.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Figure 3Description of the REAP Algorithm Comparing Exon Array Signal Estimates from hCNS-SCns and Cyt-ES(A) Histogram of Pearson correlation coefficients computed from median signal estimates for probesets between Cyt-ES versus hCNS-SCns for genes (blue bars).	[ { "end": 221, "label": "CellLine", "start": 215 }, { "end": 238, "label": "CellType", "start": 229 }, { "end": 105, "label": "CellLine", "start": 99 }, { "end": 94, "label": "CellType", "start": 85 }, { "end": 236, "label": "CellType", "sta...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Genes were required to have more than five probesets localized within the exons in the gene.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Genes", "were", "required", "to", "have", "more", ...
Red bars represented Pearson correlation coefficients computed from exons with shuffled signal estimates.(B) Each probeset contained probeset-level estimates from three replicates each, (a, b, c) in Cyt-ES and (d, e, f) in hCNS-SCns.	[ { "end": 230, "label": "CellType", "start": 223 }, { "end": 205, "label": "CellLine", "start": 199 }, { "end": 232, "label": "CellType", "start": 223 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
The five points summarizing the log2 probeset-level estimates are indicated by black filled circles.(C) Each probeset was summarized by five points.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "The", "five"...
Scatter plots of signal estimates for probesets that were present in at least one cell type (Cyt-ES or hCNS-SCns) for the EHBP1 gene.	[ { "end": 99, "label": "CellLine", "start": 93 }, { "end": 112, "label": "CellType", "start": 103 }, { "end": 110, "label": "CellType", "start": 103 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "B-CellType", "O", "O", "O", "O", "O", "...
Probesets were considered present if the DABG p-value was <0.05 for all three replicates in the cell type.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Probesets", "were", "considered", ...
A regression line derived from robust linear regression with MM estimation is indicated.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "A", "regression", "line", "derived", "from", "robust", "linear", "regressio...
Points above the line represent probesets within exons that were enriched in Cyt-ES and points below represent exons that were enriched in hCNS-SCns.	[ { "end": 83, "label": "CellLine", "start": 77 }, { "end": 148, "label": "CellType", "start": 139 }, { "end": 146, "label": "CellType", "start": 139 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O" ], "tokens"...
Points close to the regression line are not significantly different in Cyt-ES versus hCNS-SCns.	[ { "end": 94, "label": "CellType", "start": 85 }, { "end": 77, "label": "CellLine", "start": 71 }, { "end": 92, "label": "CellType", "start": 85 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "B-CellType", "O" ], "tokens": [ "Points", "close", "to", "the", "regression", "line", ...
Boxed points represented the five-point summary of a probeset that was significantly enriched in Cyt-ES but was skipped in hCNS-SCns.(D) Histogram of studentized residuals for points from the scatter plot in (C) in EHBP1.(E) The histogram of studentized residuals for all points for all analyzed probesets (100 bins).(F)...	[ { "end": 439, "label": "CellLine", "start": 433 }, { "end": 103, "label": "CellLine", "start": 97 }, { "end": 394, "label": "CellLine", "start": 388 }, { "end": 132, "label": "CellType", "start": 123 }, { "end": 423, "label": "CellType", "s...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Here, a possible representation of the data was explored.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Here", ",", "a", "possible", "representation", "of", "the", "data", "was", "explored", "."...
If we had N replicates in one condition and M replicates in the other, we could consider N*M points if we analyzed every possible pairing.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ...
For instance, three replicate signal estimates for every probeset per cell line, such as signal estimates a, b, and c in hESCs and d, e, and f in hCNS-SCns, would translate to pairing every signal (d,a), (d,b), (d,c) … (f,a), (f,b), (f,c) for linear regression (Figure 3B).	[ { "end": 155, "label": "CellType", "start": 146 }, { "end": 126, "label": "CellType", "start": 121 }, { "end": 153, "label": "CellType", "start": 146 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", ...
Instead, pairing the signal estimates of all replicates in one condition to the median of the other would only require N + M − 1 points and would capture the variation of the signal estimates of each probeset.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
For example, we considered (d,b), (e,a), (e,b), (e,c), and (f,b) points where b and d were the median intensities for the replicates in Cyt-ES and hCNS-SCns, respectively (Figure 3B).	[ { "end": 142, "label": "CellLine", "start": 136 }, { "end": 156, "label": "CellType", "start": 147 }, { "end": 154, "label": "CellType", "start": 147 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
A scatter plot of all probesets of the EHBP1 (EH domain binding protein, RefSeq identifier NM_015252) is shown in Figure 3C in the format described.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Each probeset was represented by 5 points of log-transformed (base 2) values; and each point on the scatter plot reflected the extent of inclusion of an exon in hESCs and in hCNS-SCns (Figure 3C).A classical linear regression model could be proposed to fit the response variable yij, the log2 expression of probeset i in...	[ { "end": 470, "label": "CellType", "start": 463 }, { "end": 183, "label": "CellType", "start": 174 }, { "end": 181, "label": "CellType", "start": 174 }, { "end": 166, "label": "CellType", "start": 161 }, { "end": 472, "label": "CellType", "...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
However, classical linear regression by least-squares estimation is biased because the least squares predictions are strongly influenced by the outliers, leading to completely incorrect regression line estimates, masking of the outliers, and incorrect predictions of outliers.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Therefore, we applied M-estimation robust regression to estimate the line, which is less sensitive to outliers.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Therefore", ",", "we", "applied", "M-e...
Fitting was performed using an iterated, re-weighted least squares analysis.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Fitting", "was", "performed", "using", "an", "iterated", ",", "re-weighted", "least", ...
Our assumption was that most of the points were “correct,” i.e., that most of the exons were constitutively spliced.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Our", ...
Thus, robust regression would find the line that was least dependent on outliers, which would be potential AS exons.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Thus", ",", "robu...
This assumption was substantiated by our observation that, using publicly available ESTs and mRNAs, a minority of human exons (7%) have evidence for exon-skipping, the most common form of AS.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Using robust regression, the regression line for Cyt-ESC versus hCNS-SCns in the EHBP1 gene is illustrated in Figure 3C.	[ { "end": 71, "label": "CellType", "start": 64 }, { "end": 56, "label": "CellLine", "start": 49 }, { "end": 73, "label": "CellType", "start": 64 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Using", "robust"...
The boxed points belonged to a probeset that was enriched in hESCs but depleted in hCNS-SCns, which was suspected to be due to AS.	[ { "end": 90, "label": "CellType", "start": 83 }, { "end": 92, "label": "CellType", "start": 83 }, { "end": 66, "label": "CellType", "start": 61 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "O", "O", "O", "...
The difference between the actual and regression-based predicted value, normalized by the estimate of its standard deviation, is called the studentized residuals.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ ...
Studentized residuals were computed for all probeset pairs in EHBP1, and the histogram depicting their distribution is illustrated in Figure 3D.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Studentized", ...
As expected, the mean of the distribution was close to zero, and the distribution was approximated by a t-distribution with n-p-1 degrees of freedom, where n was the number of points on the scatter plot, and the number of parameters p was 2.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
The boxed points had studentized residuals of 1.829, 3.104, 2.634, 3.012, and 2.125 with p-values of 0.034, 0.00119, 0.00477, 0.00158, and 0.01780, respectively, computed based on the t-distribution (Figure 3C).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
At a stringent p-value cutoff of 0.01, four of the five studentized residuals were designated as significant “outliers,” indicating that the probeset was “unusual.”	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
RT-PCR confirmed that the exon, represented by the probeset, was indeed differentially included in hESCs and skipped in hCNS-SCns (Figure 7B).	[ { "end": 127, "label": "CellType", "start": 120 }, { "end": 104, "label": "CellType", "start": 99 }, { "end": 129, "label": "CellType", "start": 120 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "O", "B-CellType", "O", "O", "O", "O", "...
Applying this approach to all gene models revealed that, as expected, the majority of studentized residuals are centered at zero (Figure 3E).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ...
Thus far in the example, our analysis was based on regression of hESCs (y-axis) versus hCNS-SCns (x-axis) (Figure 3B–3D).	[ { "end": 96, "label": "CellType", "start": 87 }, { "end": 70, "label": "CellType", "start": 65 }, { "end": 94, "label": "CellType", "start": 87 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "O", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "...
However, robust regression as described was not symmetrical, i.e., parameter estimation of y as a function of x was not the same as that of x as a function of y. The negative slope revealed that probesets enriched in hESCs versus hCNS-SCns (positive valued), were expectedly depleted when hCNS-SCns was compared to hESCs...	[ { "end": 222, "label": "CellType", "start": 217 }, { "end": 320, "label": "CellType", "start": 315 }, { "end": 239, "label": "CellType", "start": 230 }, { "end": 298, "label": "CellType", "start": 289 }, { "end": 237, "label": "CellType", "...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
As our method for predicting candidate alternative exons was based on identification of outliers using robust regression, we named the method REAP.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "As", ...
Identification and Removal of False PositivesIn the process of experimentally validating our predictions, we encountered three main sources of false positives (FP) from robust regression.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
First, we identified genes with probeset signal estimates that were poorly correlated and were not amenable to our method.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "First", ",", "we", "id...
As an example, the median probeset signal estimates in hESCs and hCNS-SCns of the FIP1L1 gene (gene identifiers BC011543, AL136910) had a Pearson correlation coefficient of 0.38, and the distribution of points was not amenable to robust regression (Figure 4A).	[ { "end": 60, "label": "CellType", "start": 55 }, { "end": 74, "label": "CellType", "start": 65 }, { "end": 72, "label": "CellType", "start": 65 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "...
To avoid inappropriate application of REAP and generating false predictions, we empirically determined that a gene had to have a Pearson correlation coefficient cutoff of 0.6 before being amenable to REAP analysis.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Next, we managed two additional sources of FPs, namely “high-leverage” and “high-influence” points, which we were able to identify by computing the following metrics.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
For every point, we computed (i) the studentized residual (as described above), (ii) the influence, and (iii) the leverage (see Materials and Methods for more details).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Leverage assessed how far away a value of the independent variable was from the mean value; the farther away the observation the more leverage it had.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
The influence of a point was related to its covariance ratio: a covariance ratio larger (or smaller) than 1 implied that the point was closer (or farther) than was typical to the regression line, so removing it would hurt (or help) the accuracy of the line and would increase (or decrease) the error term variance.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Influence was computed as the absolute difference between the covariance ratio and unity.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Influence", "was", "computed", "as", "the", "absolute", "difference", "betw...
To illustrate further, a point was classified as an “outlier” if it had a large studentized residual (p < 0.01) and low leverage (boxed point “a”); as a “high-leverage” point if it had a low studentized residual and high leverage (boxed point “b”); and as a “high-influence” point if it had a high studentized residual, ...	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Points that resembled boxed point “a” were designated as potential AS events.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Points", "that", "resembled", "boxed", "point", "“", "a", "”", ...
For example, four of the five boxed points in Figure 3C were “outliers,” and RT-PCR validation indicated that the exon represented by the probeset was indeed skipped in hCNS-SCns (EHBP1, Figure 7B).	[ { "end": 176, "label": "CellType", "start": 169 }, { "end": 178, "label": "CellType", "start": 169 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Points that were “high-leverage,” such as the five points in the CLCN2 gene, were experimentally verified to be a FP (Figure 4C; unpublished data).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Points that were “high-influence,” such as the four of five boxed points in the ABCA3 gene were also experimentally verified to be a FP (Figure 4D; unpublished data).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
In conclusion, in order to reduce the FP rate, all points were evaluated according to the metrics described, and points that were significant “outliers” were considered putative AS events.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Figure 4Sources of False Positives(A) Scatter plot of points for the FIP1L1 gene and the line representing the robust regression estimate.(B) Boxed point “a” represents a significant “outlier” (with a significantly different studentized residual and low leverage).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Boxed point “b” represents a “high leverage” point (low studentized residual and a high leverage).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Boxed", "point", ...
Boxed point “c” represents a “high influence” point (high studentized residual, high leverage, and high influence).(C) Scatter plot of points for the CLCN2 gene.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Boxed points represent “high leverage” points.(D) Scatter plot of points for the ABCA3 gene.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Boxed", "points", "represent", "“", "high", ...
Boxed points represent “high influence” points.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Boxed", "points", "represent", "“", "high", "influence", "”", "points", "." ] } ]
Global Identification and Characterization of REAP[+] ExonsREAP was applied to identify AS events in NPs compared to hESCs: Cyt-NP versus Cyt-ES; HUES6-NP versus HUES6-ES; hCNS-SCns versus Cyt-ES, and hCNS-SCns versus HUES6-ES.	[ { "end": 104, "label": "CellType", "start": 101 }, { "end": 122, "label": "CellType", "start": 117 }, { "end": 170, "label": "CellLine", "start": 162 }, { "end": 195, "label": "CellLine", "start": 189 }, { "end": 144, "label": "CellLine", "...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "B-CellType", "O", "O", "O", "B-CellLine", "O", "O"...
After removing potential FPs, 11,348 genes containing 158,657 probesets were scored by REAP.As described above, for each pair of cell lines compared, each probset was represented by five points, where each point was defined a significant outlier if it had a high residual (p < 0.01), low influence, and high leverage.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Points per probeset should be correlated; in other words, if one point was a significant outlier, the other points were expected to be outliers as well.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
To ensure that this was the case, we counted the number of probesets with N significant outliers, where N was varied from 0 to 5.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Next, the identity of the probesets and points derived from them were exchanged with other probesets, keeping constant the total number of points that were considered significant outliers.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
At N = 0, we observed approximately equal numbers of probesets in the actual versus shuffled controls.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "At", "N", "=", "0", ",", "we", ...
In contrast, we observed that there were 1.5 times more probesets with N = 2 significant outliers relative to shuffled controls; 12–31 times more probesets with N = 3; and 17–612 times more probesets that had N = 4 significant outliers (Figure 5A; see Table S1).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
For example, in hCNS-SCns compared to Cyt-ES, approximately 0.39% (490 of 124,604) of probesets had three significant outliers and 0.25% (308 probesets) had four significant outliers, relative to 0.02% and 0% of shuffled controls, respectively.	[ { "end": 44, "label": "CellLine", "start": 38 }, { "end": 25, "label": "CellType", "start": 16 }, { "end": 23, "label": "CellType", "start": 16 } ]	17967047	[ { "tags": [ "O", "O", "O", "O", "B-CellType", "O", "O", "B-CellLine", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "...
Figure 5Correlation between “Outliers”(A) The number of probesets with N significant “outliers” was determined for hCNS-SCns versus Cyt-ES, hCNS-SCns versus HUES6-ES, Cyt-NPs versus Cyt-ES, and HUES6-NPs versus HUES6-ES (N = 0, 1, 2, 3, 4, 5).	[ { "end": 162, "label": "CellLine", "start": 157 }, { "end": 149, "label": "CellType", "start": 140 }, { "end": 124, "label": "CellType", "start": 115 }, { "end": 165, "label": "CellLine", "start": 157 }, { "end": 138, "label": "CellLine", "...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellLine", "O", "B-CellType", "O"...
For comparison, points to probeset relationships were randomly permuted, retaining the same number of “outliers.”	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "For", "comparison", ",", "points"...
Vertical bars represent the ratio between the number of actual points and the randomly permutated sets.(B) Similar to (A), except points were counted as “outliers” only if they were “outliers” in both hCNS-SCns versus Cyt-ES and hCNS-SCns versus HUES6-ES (combined hCNS-SCns versus hESC; blue bars); in both HUES6-NP ver...	[ { "end": 238, "label": "CellType", "start": 229 }, { "end": 210, "label": "CellType", "start": 201 }, { "end": 254, "label": "CellLine", "start": 246 }, { "end": 224, "label": "CellLine", "start": 218 }, { "end": 274, "label": "CellType", "...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Comparing the significant probesets between hCNS-SCns versus Cyt-ES and hCNS-SCns versus HUES6-ES revealed 672 significant probesets (N ≥ 2), whereas if we shuffled the associations between probeset identity and significant outliers, only four significant probesets (N ≥ 2) were identified—a 168-fold enrichment (Figure ...	[ { "end": 67, "label": "CellLine", "start": 61 }, { "end": 79, "label": "CellType", "start": 72 }, { "end": 94, "label": "CellLine", "start": 89 }, { "end": 51, "label": "CellType", "start": 44 }, { "end": 81, "label": "CellType", "start": 7...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellLine", "O", "B-CellType", "O", "B-CellLine", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
A total of 236 significant probesets overlapped when we compared the derived NPs to hESCs (Cyt-NP versus Cyt-ES and HUES6-NP versus HUES6-ES), relative to seven significant probesets (34-fold enrichment).At a cutoff of two significant outliers, 1,737 probesets contained in internal exons were defined as positive REAP p...	[ { "end": 137, "label": "CellLine", "start": 132 }, { "end": 121, "label": "CellLine", "start": 116 }, { "end": 89, "label": "CellType", "start": 84 }, { "end": 80, "label": "CellType", "start": 77 }, { "end": 111, "label": "CellLine", "star...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellType", "O", "O", "O", "B-CellLine", "O", "O", "O", "B-CellLine", "O", ...
Surprisingly, we observed that the majority of REAP[+] exons were specific to the pair of hESC and NP that was compared, likely reflecting differences in genetic origins and/or culturing and differentiation conditions of the cell lines: 614 REAP[+] events were unique to hCNS-SCns versus HUE6-ES; 220 were unique to hCNS...	[ { "end": 339, "label": "CellLine", "start": 333 }, { "end": 384, "label": "CellLine", "start": 376 }, { "end": 365, "label": "CellLine", "start": 360 }, { "end": 381, "label": "CellLine", "start": 376 }, { "end": 323, "label": "CellType", "...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
The shared events between pairs of comparisons made up a minority of the total number identified: 102 REAP[+] events were found to be in common between hCNS-SCns versus Cyt-ES and hCNS-SCns versus HUES6-ES; 48 between hCNS-SCns versus HUES6-ES and HUES6-NP versus HUES6-ES; and only 17 between hCNS-SCns versus Cyt-ES an...	[ { "end": 243, "label": "CellLine", "start": 235 }, { "end": 303, "label": "CellType", "start": 294 }, { "end": 227, "label": "CellType", "start": 218 }, { "end": 175, "label": "CellLine", "start": 169 }, { "end": 161, "label": "CellType", "...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Comparison of REAP to EST-Based Method and ACEScanTraditionally, AS exons were discovered by using EST alignments to genomic loci, and also more recently by computational algorithms that used sequence information extracted from multiple genomes.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Here, we compared REAP predictions to both approaches.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Here", ",", "we", "compared", "REAP", "predictions", "to", "both", "approaches", "." ] } ]
In the first comparison, publicly available ESTs and mRNA transcripts were aligned to the human genome sequence.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "In", "the", "first", "comparison", ","...
13,934 exons with evidence for exon-skipping and/or inclusion (EST-SE for EST-verified skipped exons) were generated, comprising ∼7% of all internal exons.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tok...
First we analyzed Cyt-ES versus hCNS-SCns.	[ { "end": 24, "label": "CellLine", "start": 18 }, { "end": 41, "label": "CellType", "start": 32 }, { "end": 39, "label": "CellType", "start": 32 } ]	17967047	[ { "tags": [ "O", "O", "O", "B-CellLine", "O", "B-CellType", "O" ], "tokens": [ "First", "we", "analyzed", "Cyt-ES", "versus", "hCNS-SCns", "." ] } ]
If we required that none of the points per probeset (exon) was significant, 6% (4,402 of 71,731) of exons (after probeset mapping) had evidence for EST-SE (Figure 6A).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Shuffling the mapping between these probesets and exons resulted in 8% (5,777 of 71,731) of exons with evidence for EST-SE (Figure 6A).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
These percentages were not significantly different from the 7% of exons with EST evidence for AS observed from using all exons.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "These", "per...
By raising the requirement that probesets had to contain at least one significant point to five significant points, the percentage of EST-SE increased dramatically from 11% (531 of 4,898 exons) to 26% (33 of 126).	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
In comparison, the shuffled probesets at the same requirements remained at ∼8%, rising slightly to 11% at five points, due to small sample sizes.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
Similar trends were observed with hCNS-SCns versus HUES6-ES and the derived NPs versus hESCs (Figure 6A).	[ { "end": 43, "label": "CellType", "start": 34 }, { "end": 79, "label": "CellType", "start": 76 }, { "end": 92, "label": "CellType", "start": 87 }, { "end": 41, "label": "CellType", "start": 34 }, { "end": 56, "label": "CellLine", "start": 5...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellLine", "O", "O", "O", "B-CellType", "O", "B-CellType", "O", "O", "O", "O", "O" ], "tokens": [ "Similar", "trends", ...
Therefore, we concluded that REAP[+] exons were enriched for AS events independently identified by a transcript-based approach.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Therefore", ",", "we", "concluded...
Figure 6Comparison of REAP Predictions for hCNS-SCns versus Cyt-hES, hCNS-SCns versus HUES6-ES, Cyt-NP versus Cyt-ES, and HUES6-NPs versus HUES6-ES with Alternative Exons Identified by an EST-Based Method and ACEScan(A) Black-filled squares represented the fraction of exons containing probesets with N significant point...	[ { "end": 67, "label": "CellLine", "start": 60 }, { "end": 78, "label": "CellType", "start": 69 }, { "end": 127, "label": "CellLine", "start": 122 }, { "end": 144, "label": "CellLine", "start": 139 }, { "end": 131, "label": "CellType", "star...	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellLine", "O", "B-CellType", "O", "B-CellLine", "O", "O", "O", "B-CellLine", "O", "O", "B-CellType", "O", "B-CellLi...
White-filled triangles represented similarly computed fractions with permuted probeset to exon mappings.(B) Black-filled squares represented the fraction of exons containing probesets with N significant points that had ACEScan positive scores, indicative of evolutionarily conserved alternative exons.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...
White-filled triangles represented similarly computed fractions with permuted probeset to exon mappings.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "White-filled", "triangles", "represented", "similarly", "computed", "fractions", "with", ...
Next, we compared REAP predictions to a computational approach of identifying exons with AS conserved in human and mouse, ACEScan [55].	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tok...
ACEScan receives as input orthologous human–mouse exon pairs and flanking intronic regions and computes sequence features and integrates the features into a machine-learning algorithm to assign a real-valued score to the exon.	[]	17967047	[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

GO categories are ordered from top to bottom in order of decreasingly significant bias toward enriched genes.(A) GO analysis of enriched genes in hESCs.(B) GO analysis of enriched genes in NPs.

[ { "end": 192, "label": "CellType", "start": 189 }, { "end": 151, "label": "CellType", "start": 146 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

To summarize, firstly immunohistochemical and RT-PCR evidence validated that the cells exhibited expected characteristics; secondly, stage-specific marker gene differences by RT-PCR were reflected accurately by gene-level estimates from the exon arrays; thirdly, the hESC-enriched genes were coherent with known genes th...

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Description of the Regression-Based Exon Array ProtocolConvinced that the signal estimates from the exon arrays reflected expected molecular and biological differences between hESCs and NPs, we sought to identify AS events.

[ { "end": 189, "label": "CellType", "start": 186 }, { "end": 181, "label": "CellType", "start": 176 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellType", "...

We compared Cyt-ES to hCNS-SCns to illustrate our approach.

[ { "end": 29, "label": "CellType", "start": 22 }, { "end": 18, "label": "CellLine", "start": 12 }, { "end": 31, "label": "CellType", "start": 22 } ]

17967047

[ { "tags": [ "O", "O", "B-CellLine", "O", "B-CellType", "O", "O", "O", "O", "O" ], "tokens": [ "We", "compared", "Cyt-ES", "to", "hCNS-SCns", "to", "illustrate", "our", "approach", "." ...

First we normalized the data and generated signal estimates with Robust Multichip Analysis (RMA) and estimated the probability that each probeset was detected above background (DABG) using publicly available Affymetrix Power Tools (APT).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

We analyzed probesets that (i) comprised three or more individual probes; (ii) were localized within the exons of our gene models with evidence from at least three sources (mRNA, EST, or full-length cDNA); and (iii) were detected above background in at least one of the cell lines.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

In total, 17,430 gene models were represented by probesets that satisfied these criteria.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "In", "total", ",", "17,430", "gene", "models", "were", "represen...

Next we asked whether the probeset expression within each gene model was positively correlated for any two cell lines.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Next", "we", "asked", "whether", ...

To do this we calculated the Pearson correlation coefficient between the vectors of median signal estimates across replicates in Cyt-ES versus hCNS-SCns.

[ { "end": 150, "label": "CellType", "start": 143 }, { "end": 152, "label": "CellType", "start": 143 }, { "end": 135, "label": "CellLine", "start": 129 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "B-CellType", "O" ], "tokens": [ "...

The vast majority of genes (>80%) was found to have probeset-level Pearson correlation coefficients of greater than 0.8 (Figure 3A).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ...

Next we randomly permuted the association between the median signal estimates and the probesets for each gene in hESCs (or hCNS-SCns) and observed that the distribution of Pearson correlation coefficients for the permuted sets was centered at zero, as expected (Figure 3A).

[ { "end": 130, "label": "CellType", "start": 123 }, { "end": 132, "label": "CellType", "start": 123 }, { "end": 118, "label": "CellType", "start": 113 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "B-CellType", "O", "O", "O", "...

This indicated that the signal estimates for probesets between hESCs and hCNS-SCns were highly correlated and suggested that a scatter plot of probeset signal estimates between hESCs and hCNS-SCns would reveal a linear relationship for the majority of genes.

[ { "end": 182, "label": "CellType", "start": 177 }, { "end": 82, "label": "CellType", "start": 73 }, { "end": 196, "label": "CellType", "start": 187 }, { "end": 68, "label": "CellType", "start": 63 }, { "end": 194, "label": "CellType", "star...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "...

We hypothesized that a linear regression to determine if some probesets behaved unexpectedly in one cell type compared to the other might be a reasonable approach to identify AS exons.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Figure 3Description of the REAP Algorithm Comparing Exon Array Signal Estimates from hCNS-SCns and Cyt-ES(A) Histogram of Pearson correlation coefficients computed from median signal estimates for probesets between Cyt-ES versus hCNS-SCns for genes (blue bars).

[ { "end": 221, "label": "CellLine", "start": 215 }, { "end": 238, "label": "CellType", "start": 229 }, { "end": 105, "label": "CellLine", "start": 99 }, { "end": 94, "label": "CellType", "start": 85 }, { "end": 236, "label": "CellType", "sta...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Genes were required to have more than five probesets localized within the exons in the gene.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Genes", "were", "required", "to", "have", "more", ...

Red bars represented Pearson correlation coefficients computed from exons with shuffled signal estimates.(B) Each probeset contained probeset-level estimates from three replicates each, (a, b, c) in Cyt-ES and (d, e, f) in hCNS-SCns.

[ { "end": 230, "label": "CellType", "start": 223 }, { "end": 205, "label": "CellLine", "start": 199 }, { "end": 232, "label": "CellType", "start": 223 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

The five points summarizing the log2 probeset-level estimates are indicated by black filled circles.(C) Each probeset was summarized by five points.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "The", "five"...

Scatter plots of signal estimates for probesets that were present in at least one cell type (Cyt-ES or hCNS-SCns) for the EHBP1 gene.

[ { "end": 99, "label": "CellLine", "start": 93 }, { "end": 112, "label": "CellType", "start": 103 }, { "end": 110, "label": "CellType", "start": 103 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "B-CellType", "O", "O", "O", "O", "O", "...

Probesets were considered present if the DABG p-value was <0.05 for all three replicates in the cell type.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Probesets", "were", "considered", ...

A regression line derived from robust linear regression with MM estimation is indicated.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "A", "regression", "line", "derived", "from", "robust", "linear", "regressio...

Points above the line represent probesets within exons that were enriched in Cyt-ES and points below represent exons that were enriched in hCNS-SCns.

[ { "end": 83, "label": "CellLine", "start": 77 }, { "end": 148, "label": "CellType", "start": 139 }, { "end": 146, "label": "CellType", "start": 139 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O" ], "tokens"...

Points close to the regression line are not significantly different in Cyt-ES versus hCNS-SCns.

[ { "end": 94, "label": "CellType", "start": 85 }, { "end": 77, "label": "CellLine", "start": 71 }, { "end": 92, "label": "CellType", "start": 85 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "B-CellType", "O" ], "tokens": [ "Points", "close", "to", "the", "regression", "line", ...

Boxed points represented the five-point summary of a probeset that was significantly enriched in Cyt-ES but was skipped in hCNS-SCns.(D) Histogram of studentized residuals for points from the scatter plot in (C) in EHBP1.(E) The histogram of studentized residuals for all points for all analyzed probesets (100 bins).(F)...

[ { "end": 439, "label": "CellLine", "start": 433 }, { "end": 103, "label": "CellLine", "start": 97 }, { "end": 394, "label": "CellLine", "start": 388 }, { "end": 132, "label": "CellType", "start": 123 }, { "end": 423, "label": "CellType", "s...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Here, a possible representation of the data was explored.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Here", ",", "a", "possible", "representation", "of", "the", "data", "was", "explored", "."...

If we had N replicates in one condition and M replicates in the other, we could consider N*M points if we analyzed every possible pairing.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ...

For instance, three replicate signal estimates for every probeset per cell line, such as signal estimates a, b, and c in hESCs and d, e, and f in hCNS-SCns, would translate to pairing every signal (d,a), (d,b), (d,c) … (f,a), (f,b), (f,c) for linear regression (Figure 3B).

[ { "end": 155, "label": "CellType", "start": 146 }, { "end": 126, "label": "CellType", "start": 121 }, { "end": 153, "label": "CellType", "start": 146 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", ...

Instead, pairing the signal estimates of all replicates in one condition to the median of the other would only require N + M − 1 points and would capture the variation of the signal estimates of each probeset.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

For example, we considered (d,b), (e,a), (e,b), (e,c), and (f,b) points where b and d were the median intensities for the replicates in Cyt-ES and hCNS-SCns, respectively (Figure 3B).

[ { "end": 142, "label": "CellLine", "start": 136 }, { "end": 156, "label": "CellType", "start": 147 }, { "end": 154, "label": "CellType", "start": 147 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

A scatter plot of all probesets of the EHBP1 (EH domain binding protein, RefSeq identifier NM_015252) is shown in Figure 3C in the format described.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Each probeset was represented by 5 points of log-transformed (base 2) values; and each point on the scatter plot reflected the extent of inclusion of an exon in hESCs and in hCNS-SCns (Figure 3C).A classical linear regression model could be proposed to fit the response variable yij, the log2 expression of probeset i in...

[ { "end": 470, "label": "CellType", "start": 463 }, { "end": 183, "label": "CellType", "start": 174 }, { "end": 181, "label": "CellType", "start": 174 }, { "end": 166, "label": "CellType", "start": 161 }, { "end": 472, "label": "CellType", "...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

However, classical linear regression by least-squares estimation is biased because the least squares predictions are strongly influenced by the outliers, leading to completely incorrect regression line estimates, masking of the outliers, and incorrect predictions of outliers.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Therefore, we applied M-estimation robust regression to estimate the line, which is less sensitive to outliers.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Therefore", ",", "we", "applied", "M-e...

Fitting was performed using an iterated, re-weighted least squares analysis.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Fitting", "was", "performed", "using", "an", "iterated", ",", "re-weighted", "least", ...

Our assumption was that most of the points were “correct,” i.e., that most of the exons were constitutively spliced.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Our", ...

Thus, robust regression would find the line that was least dependent on outliers, which would be potential AS exons.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Thus", ",", "robu...

This assumption was substantiated by our observation that, using publicly available ESTs and mRNAs, a minority of human exons (7%) have evidence for exon-skipping, the most common form of AS.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Using robust regression, the regression line for Cyt-ESC versus hCNS-SCns in the EHBP1 gene is illustrated in Figure 3C.

[ { "end": 71, "label": "CellType", "start": 64 }, { "end": 56, "label": "CellLine", "start": 49 }, { "end": 73, "label": "CellType", "start": 64 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "B-CellLine", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Using", "robust"...

The boxed points belonged to a probeset that was enriched in hESCs but depleted in hCNS-SCns, which was suspected to be due to AS.

[ { "end": 90, "label": "CellType", "start": 83 }, { "end": 92, "label": "CellType", "start": 83 }, { "end": 66, "label": "CellType", "start": 61 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "O", "O", "O", "...

The difference between the actual and regression-based predicted value, normalized by the estimate of its standard deviation, is called the studentized residuals.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ ...

Studentized residuals were computed for all probeset pairs in EHBP1, and the histogram depicting their distribution is illustrated in Figure 3D.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Studentized", ...

As expected, the mean of the distribution was close to zero, and the distribution was approximated by a t-distribution with n-p-1 degrees of freedom, where n was the number of points on the scatter plot, and the number of parameters p was 2.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

The boxed points had studentized residuals of 1.829, 3.104, 2.634, 3.012, and 2.125 with p-values of 0.034, 0.00119, 0.00477, 0.00158, and 0.01780, respectively, computed based on the t-distribution (Figure 3C).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

At a stringent p-value cutoff of 0.01, four of the five studentized residuals were designated as significant “outliers,” indicating that the probeset was “unusual.”

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

RT-PCR confirmed that the exon, represented by the probeset, was indeed differentially included in hESCs and skipped in hCNS-SCns (Figure 7B).

[ { "end": 127, "label": "CellType", "start": 120 }, { "end": 104, "label": "CellType", "start": 99 }, { "end": 129, "label": "CellType", "start": 120 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "O", "B-CellType", "O", "O", "O", "O", "...

Applying this approach to all gene models revealed that, as expected, the majority of studentized residuals are centered at zero (Figure 3E).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ...

Thus far in the example, our analysis was based on regression of hESCs (y-axis) versus hCNS-SCns (x-axis) (Figure 3B–3D).

[ { "end": 96, "label": "CellType", "start": 87 }, { "end": 70, "label": "CellType", "start": 65 }, { "end": 94, "label": "CellType", "start": 87 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "O", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "...

However, robust regression as described was not symmetrical, i.e., parameter estimation of y as a function of x was not the same as that of x as a function of y. The negative slope revealed that probesets enriched in hESCs versus hCNS-SCns (positive valued), were expectedly depleted when hCNS-SCns was compared to hESCs...

[ { "end": 222, "label": "CellType", "start": 217 }, { "end": 320, "label": "CellType", "start": 315 }, { "end": 239, "label": "CellType", "start": 230 }, { "end": 298, "label": "CellType", "start": 289 }, { "end": 237, "label": "CellType", "...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

As our method for predicting candidate alternative exons was based on identification of outliers using robust regression, we named the method REAP.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "As", ...

Identification and Removal of False PositivesIn the process of experimentally validating our predictions, we encountered three main sources of false positives (FP) from robust regression.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

First, we identified genes with probeset signal estimates that were poorly correlated and were not amenable to our method.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "First", ",", "we", "id...

As an example, the median probeset signal estimates in hESCs and hCNS-SCns of the FIP1L1 gene (gene identifiers BC011543, AL136910) had a Pearson correlation coefficient of 0.38, and the distribution of points was not amenable to robust regression (Figure 4A).

[ { "end": 60, "label": "CellType", "start": 55 }, { "end": 74, "label": "CellType", "start": 65 }, { "end": 72, "label": "CellType", "start": 65 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellType", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "...

To avoid inappropriate application of REAP and generating false predictions, we empirically determined that a gene had to have a Pearson correlation coefficient cutoff of 0.6 before being amenable to REAP analysis.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Next, we managed two additional sources of FPs, namely “high-leverage” and “high-influence” points, which we were able to identify by computing the following metrics.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

For every point, we computed (i) the studentized residual (as described above), (ii) the influence, and (iii) the leverage (see Materials and Methods for more details).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Leverage assessed how far away a value of the independent variable was from the mean value; the farther away the observation the more leverage it had.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

The influence of a point was related to its covariance ratio: a covariance ratio larger (or smaller) than 1 implied that the point was closer (or farther) than was typical to the regression line, so removing it would hurt (or help) the accuracy of the line and would increase (or decrease) the error term variance.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Influence was computed as the absolute difference between the covariance ratio and unity.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Influence", "was", "computed", "as", "the", "absolute", "difference", "betw...

To illustrate further, a point was classified as an “outlier” if it had a large studentized residual (p < 0.01) and low leverage (boxed point “a”); as a “high-leverage” point if it had a low studentized residual and high leverage (boxed point “b”); and as a “high-influence” point if it had a high studentized residual, ...

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Points that resembled boxed point “a” were designated as potential AS events.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Points", "that", "resembled", "boxed", "point", "“", "a", "”", ...

For example, four of the five boxed points in Figure 3C were “outliers,” and RT-PCR validation indicated that the exon represented by the probeset was indeed skipped in hCNS-SCns (EHBP1, Figure 7B).

[ { "end": 176, "label": "CellType", "start": 169 }, { "end": 178, "label": "CellType", "start": 169 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Points that were “high-leverage,” such as the five points in the CLCN2 gene, were experimentally verified to be a FP (Figure 4C; unpublished data).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Points that were “high-influence,” such as the four of five boxed points in the ABCA3 gene were also experimentally verified to be a FP (Figure 4D; unpublished data).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

In conclusion, in order to reduce the FP rate, all points were evaluated according to the metrics described, and points that were significant “outliers” were considered putative AS events.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Figure 4Sources of False Positives(A) Scatter plot of points for the FIP1L1 gene and the line representing the robust regression estimate.(B) Boxed point “a” represents a significant “outlier” (with a significantly different studentized residual and low leverage).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Boxed point “b” represents a “high leverage” point (low studentized residual and a high leverage).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Boxed", "point", ...

Boxed point “c” represents a “high influence” point (high studentized residual, high leverage, and high influence).(C) Scatter plot of points for the CLCN2 gene.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Boxed points represent “high leverage” points.(D) Scatter plot of points for the ABCA3 gene.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Boxed", "points", "represent", "“", "high", ...

Boxed points represent “high influence” points.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Boxed", "points", "represent", "“", "high", "influence", "”", "points", "." ] } ]

Global Identification and Characterization of REAP[+] ExonsREAP was applied to identify AS events in NPs compared to hESCs: Cyt-NP versus Cyt-ES; HUES6-NP versus HUES6-ES; hCNS-SCns versus Cyt-ES, and hCNS-SCns versus HUES6-ES.

[ { "end": 104, "label": "CellType", "start": 101 }, { "end": 122, "label": "CellType", "start": 117 }, { "end": 170, "label": "CellLine", "start": 162 }, { "end": 195, "label": "CellLine", "start": 189 }, { "end": 144, "label": "CellLine", "...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "O", "B-CellType", "O", "O", "O", "B-CellLine", "O", "O"...

After removing potential FPs, 11,348 genes containing 158,657 probesets were scored by REAP.As described above, for each pair of cell lines compared, each probset was represented by five points, where each point was defined a significant outlier if it had a high residual (p < 0.01), low influence, and high leverage.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Points per probeset should be correlated; in other words, if one point was a significant outlier, the other points were expected to be outliers as well.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

To ensure that this was the case, we counted the number of probesets with N significant outliers, where N was varied from 0 to 5.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Next, the identity of the probesets and points derived from them were exchanged with other probesets, keeping constant the total number of points that were considered significant outliers.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

At N = 0, we observed approximately equal numbers of probesets in the actual versus shuffled controls.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "At", "N", "=", "0", ",", "we", ...

In contrast, we observed that there were 1.5 times more probesets with N = 2 significant outliers relative to shuffled controls; 12–31 times more probesets with N = 3; and 17–612 times more probesets that had N = 4 significant outliers (Figure 5A; see Table S1).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

For example, in hCNS-SCns compared to Cyt-ES, approximately 0.39% (490 of 124,604) of probesets had three significant outliers and 0.25% (308 probesets) had four significant outliers, relative to 0.02% and 0% of shuffled controls, respectively.

[ { "end": 44, "label": "CellLine", "start": 38 }, { "end": 25, "label": "CellType", "start": 16 }, { "end": 23, "label": "CellType", "start": 16 } ]

17967047

[ { "tags": [ "O", "O", "O", "O", "B-CellType", "O", "O", "B-CellLine", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "...

Figure 5Correlation between “Outliers”(A) The number of probesets with N significant “outliers” was determined for hCNS-SCns versus Cyt-ES, hCNS-SCns versus HUES6-ES, Cyt-NPs versus Cyt-ES, and HUES6-NPs versus HUES6-ES (N = 0, 1, 2, 3, 4, 5).

[ { "end": 162, "label": "CellLine", "start": 157 }, { "end": 149, "label": "CellType", "start": 140 }, { "end": 124, "label": "CellType", "start": 115 }, { "end": 165, "label": "CellLine", "start": 157 }, { "end": 138, "label": "CellLine", "...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellLine", "O", "B-CellType", "O"...

For comparison, points to probeset relationships were randomly permuted, retaining the same number of “outliers.”

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "For", "comparison", ",", "points"...

Vertical bars represent the ratio between the number of actual points and the randomly permutated sets.(B) Similar to (A), except points were counted as “outliers” only if they were “outliers” in both hCNS-SCns versus Cyt-ES and hCNS-SCns versus HUES6-ES (combined hCNS-SCns versus hESC; blue bars); in both HUES6-NP ver...

[ { "end": 238, "label": "CellType", "start": 229 }, { "end": 210, "label": "CellType", "start": 201 }, { "end": 254, "label": "CellLine", "start": 246 }, { "end": 224, "label": "CellLine", "start": 218 }, { "end": 274, "label": "CellType", "...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Comparing the significant probesets between hCNS-SCns versus Cyt-ES and hCNS-SCns versus HUES6-ES revealed 672 significant probesets (N ≥ 2), whereas if we shuffled the associations between probeset identity and significant outliers, only four significant probesets (N ≥ 2) were identified—a 168-fold enrichment (Figure ...

[ { "end": 67, "label": "CellLine", "start": 61 }, { "end": 79, "label": "CellType", "start": 72 }, { "end": 94, "label": "CellLine", "start": 89 }, { "end": 51, "label": "CellType", "start": 44 }, { "end": 81, "label": "CellType", "start": 7...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellLine", "O", "B-CellType", "O", "B-CellLine", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

A total of 236 significant probesets overlapped when we compared the derived NPs to hESCs (Cyt-NP versus Cyt-ES and HUES6-NP versus HUES6-ES), relative to seven significant probesets (34-fold enrichment).At a cutoff of two significant outliers, 1,737 probesets contained in internal exons were defined as positive REAP p...

[ { "end": 137, "label": "CellLine", "start": 132 }, { "end": 121, "label": "CellLine", "start": 116 }, { "end": 89, "label": "CellType", "start": 84 }, { "end": 80, "label": "CellType", "start": 77 }, { "end": 111, "label": "CellLine", "star...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellType", "O", "O", "O", "B-CellLine", "O", "O", "O", "B-CellLine", "O", ...

Surprisingly, we observed that the majority of REAP[+] exons were specific to the pair of hESC and NP that was compared, likely reflecting differences in genetic origins and/or culturing and differentiation conditions of the cell lines: 614 REAP[+] events were unique to hCNS-SCns versus HUE6-ES; 220 were unique to hCNS...

[ { "end": 339, "label": "CellLine", "start": 333 }, { "end": 384, "label": "CellLine", "start": 376 }, { "end": 365, "label": "CellLine", "start": 360 }, { "end": 381, "label": "CellLine", "start": 376 }, { "end": 323, "label": "CellType", "...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

The shared events between pairs of comparisons made up a minority of the total number identified: 102 REAP[+] events were found to be in common between hCNS-SCns versus Cyt-ES and hCNS-SCns versus HUES6-ES; 48 between hCNS-SCns versus HUES6-ES and HUES6-NP versus HUES6-ES; and only 17 between hCNS-SCns versus Cyt-ES an...

[ { "end": 243, "label": "CellLine", "start": 235 }, { "end": 303, "label": "CellType", "start": 294 }, { "end": 227, "label": "CellType", "start": 218 }, { "end": 175, "label": "CellLine", "start": 169 }, { "end": 161, "label": "CellType", "...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Comparison of REAP to EST-Based Method and ACEScanTraditionally, AS exons were discovered by using EST alignments to genomic loci, and also more recently by computational algorithms that used sequence information extracted from multiple genomes.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Here, we compared REAP predictions to both approaches.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Here", ",", "we", "compared", "REAP", "predictions", "to", "both", "approaches", "." ] } ]

In the first comparison, publicly available ESTs and mRNA transcripts were aligned to the human genome sequence.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "In", "the", "first", "comparison", ","...

13,934 exons with evidence for exon-skipping and/or inclusion (EST-SE for EST-verified skipped exons) were generated, comprising ∼7% of all internal exons.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tok...

First we analyzed Cyt-ES versus hCNS-SCns.

[ { "end": 24, "label": "CellLine", "start": 18 }, { "end": 41, "label": "CellType", "start": 32 }, { "end": 39, "label": "CellType", "start": 32 } ]

17967047

[ { "tags": [ "O", "O", "O", "B-CellLine", "O", "B-CellType", "O" ], "tokens": [ "First", "we", "analyzed", "Cyt-ES", "versus", "hCNS-SCns", "." ] } ]

If we required that none of the points per probeset (exon) was significant, 6% (4,402 of 71,731) of exons (after probeset mapping) had evidence for EST-SE (Figure 6A).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Shuffling the mapping between these probesets and exons resulted in 8% (5,777 of 71,731) of exons with evidence for EST-SE (Figure 6A).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

These percentages were not significantly different from the 7% of exons with EST evidence for AS observed from using all exons.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "These", "per...

By raising the requirement that probesets had to contain at least one significant point to five significant points, the percentage of EST-SE increased dramatically from 11% (531 of 4,898 exons) to 26% (33 of 126).

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

In comparison, the shuffled probesets at the same requirements remained at ∼8%, rising slightly to 11% at five points, due to small sample sizes.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

Similar trends were observed with hCNS-SCns versus HUES6-ES and the derived NPs versus hESCs (Figure 6A).

[ { "end": 43, "label": "CellType", "start": 34 }, { "end": 79, "label": "CellType", "start": 76 }, { "end": 92, "label": "CellType", "start": 87 }, { "end": 41, "label": "CellType", "start": 34 }, { "end": 56, "label": "CellLine", "start": 5...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellLine", "O", "O", "O", "B-CellType", "O", "B-CellType", "O", "O", "O", "O", "O" ], "tokens": [ "Similar", "trends", ...

Therefore, we concluded that REAP[+] exons were enriched for AS events independently identified by a transcript-based approach.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "Therefore", ",", "we", "concluded...

Figure 6Comparison of REAP Predictions for hCNS-SCns versus Cyt-hES, hCNS-SCns versus HUES6-ES, Cyt-NP versus Cyt-ES, and HUES6-NPs versus HUES6-ES with Alternative Exons Identified by an EST-Based Method and ACEScan(A) Black-filled squares represented the fraction of exons containing probesets with N significant point...

[ { "end": 67, "label": "CellLine", "start": 60 }, { "end": 78, "label": "CellType", "start": 69 }, { "end": 127, "label": "CellLine", "start": 122 }, { "end": 144, "label": "CellLine", "start": 139 }, { "end": 131, "label": "CellType", "star...

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "B-CellType", "O", "B-CellLine", "O", "B-CellType", "O", "B-CellLine", "O", "O", "O", "B-CellLine", "O", "O", "B-CellType", "O", "B-CellLi...

White-filled triangles represented similarly computed fractions with permuted probeset to exon mappings.(B) Black-filled squares represented the fraction of exons containing probesets with N significant points that had ACEScan positive scores, indicative of evolutionarily conserved alternative exons.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...

White-filled triangles represented similarly computed fractions with permuted probeset to exon mappings.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tokens": [ "White-filled", "triangles", "represented", "similarly", "computed", "fractions", "with", ...

Next, we compared REAP predictions to a computational approach of identifying exons with AS conserved in human and mouse, ACEScan [55].

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O" ], "tok...

ACEScan receives as input orthologous human–mouse exon pairs and flanking intronic regions and computes sequence features and integrates the features into a machine-learning algorithm to assign a real-valued score to the exon.

[]

17967047

[ { "tags": [ "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", "O", ...