Browsed by
Author: Andrew Walsh

“Phenoville” – RNAi & CRISPR Screening Strategies

“Phenoville” – RNAi & CRISPR Screening Strategies

Pleasantville is a movie based on an interesting idea: two teenagers are magically transported through their TV to a town called Pleasantville set in the 1950s where everything is perfect (and also black-and-white).  As they discover the complex, imperfect emotions hidden below the idyllic surface, the black-and-white characters and objects start to gain colour.

In loss-of-function genetic screening, some reagents and screening formats may also give rise to a narrow, black-and-white view of a biological process.  A sort of “Phenoville”.  This was illustrated nicely in a recent review of screening strategies for human-virus interactions by Perreira et al. (2016).

The authors performed screens for human rhinovirus (HRV) infection using arrayed RNAi reagents (siRNAs) and pooled CRISPR reagents (sgRNAs), and then compared the resulting hit lists.

The arrayed RNAi screen produced over 160 high-confidence candidate genes, whereas the CRISPR screen only found 2.  The authors comment:

“The comparison of these two screening approaches side-by-side, using the same cells and virus, raises an interesting point. The number of host factors found for HRV14 was far greater using the MORR/RIGER approach [i.e. RNAi performed with multiple orthologous RNAi reagents and analysed by RNAi gene enrichment ranking method] and is approaching a systems level understanding based on bioinformatic analyses and the near saturation of, or enrichment for, multiple complexes and pathways (Fig. 4) (Perreira et al., 2015). By comparison our matched pooled CRISPR/Cas9 screen for HRV-HFs yielded two high-confidence candidates based on reagent redundancy, ICAM1, the known receptor for HRV14, and EXOC4, a gene involved in exocyst targeting and vesicular transport (He & Guo, 2009). Given the known role of ICAM1 as the host receptor for most HRVs, these results point to entry as the major viral lifecycle stage interrogated by a pooled functional genomic screening approach using a population of randomly biallelic null cells infected by a cytopathic virus.”

In simple terms, RNAi screening produced a richer data set that revealed system level interactions whereas CRISPR screening yielded a small number of specific hits that only affected an early-stage pathway. The ‘systems level understanding’ is nicely shown in the following diagram of the RNAi hits.  The red box at the top left is the only gene (ICAM1) that was common to the RNAi and CRISPR screens.

Perreira et al. conclude that arrayed siRNA screens permit the detection of a larger number of viral dependency factors, albeit with a significant tradeoff in a greater number of false positive hits (mainly due to off-target effects).  In contrast, pooled screens with CRISPR sgRNAs using cell survival as a readout, as also seen with most haploid cell screens, display limited sensitivity but excellent specificity in finding host genes that act early on in viral replication (e.g. ICAM1).

In Perreira et al.‘s words:

“… given the currently available functional genomic strategies if the goal is to find viral entry factors (e.g., host receptors) with high specificity its best to use a pooled survival screen, but alternatively if the aim is to obtain with relative ease a more comprehensive set of host factors, albeit with more prevalent false positives, than an arrayed siRNA screen would be the preferred method.”

Summarizing two options for genetic screeners:

  1. Arrayed RNAi screens
    • provide a richer view of the underlying biology
    • produce more false positives from OTEs
    • produce false negatives from OTEs
  2. Pooled CRISPR screens
    • provide a narrower view of the underlying biology
    • produce fewer false positives
    • produce false negatives because of genetic compensation

Off-target effects (OTEs) are the primary cause of false positives, and the resultant higher assay noise also increases the number of false negatives in arrayed RNAi screens. Reagents like siPOOLs minimize the risk of off-target effects and reduce assay noise.

One key factor not mentioned by Perreira et al. is the presence of genetic compensation in gene knockout approaches.

Putting genetic compensation in terms of human actors, imagine that you are investigating the function of bus drivers in Pleasantville.  To induce loss-of-function, assume that aliens will be abducting the bus drivers.  If the bus drivers are abducted in their sleep (equivalent to a CRISPR knock-out), you may not get a good idea of their function when you film the next day.  People may be compensating by driving, biking or staying home.  Alternatively, the bus company may have found emergency replacement drivers.

Now suppose the bus drivers are abducted in the middle of the day while driving their routes (equivalent to an RNAi knock-down).  The film will show buses crashing (hopefully without any serious injuries, since this is just a TV show!) and the public transportation system will suddenly come to a halt.

RNAi gene knockdown screens with siPOOLs  can provide a significant advantage over CRISPR gene knockout screens in obtaining a system level understanding in biological models.

Want to receive regular blog updates? Sign up for our siTOOLs Newsletter:

Follow us or share this post:
Making sense of siGENOME deconvolution

Making sense of siGENOME deconvolution

As discussed previously, deconvoluted Dharmacon siGENOME pools often give surprising results.  (Deconvolution is the process of testing the 4 siRNAs in a pool individually.  This is usually done in the validation phase of siRNA screens.)

One way to compare the relative contribution of target gene and off-target effects is to calculate the correlation between reagents having the same target gene or the same seed sequence.  One of the first things we do when analysing single siRNA screens is to calculate a robust form of the intraclass correlation (rICC, see discussion at bottom for more about this).

Recently we were analysing deconvolution data from Adamson et al. (2012) and calculated the following rICC’s.  (The phenotype measured was relative homologous recombination.)

Grouping variable  rICC    95% confidence interval

Target gene        0.040   -0.021-0.099
Antisense 7mer     0.383   0.357-0.413
Sense 7mer         0.093   0.054-0.129

Besides the order of magnitude difference between target gene and antisense seed correlation (which is commonly observed in RNAi screens), what stands out is the ~2-fold difference between the correlation by target gene and sense seed.

Very little of the the sense strand should be loaded into RISC, if the siRNAs were designed with appropriate thermodynamic considerations (the 5′ antisense end should be less stable than the 5′ sense end, to ensure that the antisense strand is preferentially loaded into RISC).

The above correlations suggest that some not insubstantial amount of sense strand is making it into the RISC complex.

Here is the distribution of delta-delta-G for siPOOLs and siGENOME siRNAs targeting the same 500 human kinases (see bottom of post for discussion of calculation).  A positive delta-delta G means that the sense end is more thermodynamically stable than the antisense end, favouring the loading of the antisense strand into RISC.



This discrepancy in delta-delta G is also consistent with comparison of mRNA knockdown:

The siGENOME knockdown data comes from 774 genes analysed by qPCR in Simpson et al. (2008).  The siPOOL knockdown data is from 223 genes where we have done qPCR validation.

Of note, the siGENOME pools were tested at 100 nM, whereas siPOOLs were tested at 1 nM.

(It should be mentioned that, although consistent with the observed differences in ddG, this is only an indirect comparison, and delta-delta G is not the only determinant of functional siRNAs.)


Notes on intraclass correlation

Intraclass correlation measures the agreement between multiple measurements (in this case, multiple siRNAs with the same target gene, or multiple siRNAs with the same seed sequence).   One could also pair off all the repeated measures and calculate correlation using standard methods (parametrically using Pearson’s method, or non-parametrically using Spearman’s method).  The main problem with such an approach is that there is no natural way to determine which measure goes in the x or y column.  Correlations are normally between different variables (e.g. height and weight).  In a case of repeated measures, there is no natural order, so the intraclass correlation (ICC) is the more correct way to measure the similarity of within-group measurements.  As ICC depends on a normal distribution, datasets must first be examined, and if necessary, transformed beforehand.

Robust methods have the advantage of permitting the use of untransformed data, which is especially useful when running scripts across hundreds of screening dataset features.  The algorithm we use calculates a robust approximation of the ICC by combining resampling and non-parametric correlation.

Here is the algorithm, in a nutshell:

  1. Group observations (e.g. cell count) by the grouping variable (e.g. target gene or antisense seed)
  2. Randomly assign one value of each group to the x or y column (groups with one 1 observation are skipped)
    • for example, if the grouping variable is target gene and siRNAs targeting PLK1 had the values 23, 30, 37, 45, the program would randomly choose 1 of the values for the x column and another for the y column
  3. Calcule Spearman’s rho (non-parametric measure of correlation)
  4. Repeat steps 1-3 a set number of times (e.g. 300) and store the calculated rho’s
  5. Calculate mean of the rho values from 4.  This is the robust approximation of the ICC (rICC).
    • Values from 4 are also used to calculate confidence intervals.

The program that calculates this is available upon request.

Notes on calculating delta-delta G

Delta-delta G was calculated using the Vienna RNA package, as detailed here: (in answer by Brad Chapman).

The delta-delta G was calculated using 3 terminal bps.  We found that that ddG of the terminal 3 bps had the strongest correlation with observed knockdown.  Others (e.g. Schwarz et al., 2003 and Khvorova et al., 2003) have also used the terminal 4 bps.


Want to receive regular blog updates? Sign up for our siTOOLs Newsletter:

Follow us or share this post:
How reproducible are CRISPR screens?

How reproducible are CRISPR screens?

The reproducibility of different CRISPR or RNAi reagents targeting the same gene is sometimes cited as prima facie evidence for the superiority of CRISPR screens to RNAi screens.

A landmark paper by Shalem et al. showed that different gRNAs inhibit gene expression much more consistently than do different shRNAs:

But does this ensure that CRISPR screens are more reliable (as determined by assay reproducibility) than RNAi screens?  Not necessarily.

Shalem et al. performed two pooled CRISPR screens in parallel, and found substantial overlap between the top hits.

How does this overlap compare to that between replicate RNAi screens?

In 2010, Barrows et al. tested the reproducibility between genome-wide siRNA screens conducted 5 months apart.  Using the sum of ranks hit selection algorithm, they found 75 and 82 hits from the first and second screens, respectively, with 43 hits overlapping.

If we take the top 75 and top 82 hits from the Shalem replicate screens, we only find 17 genes overlapping.

It’s important to note that the Shalem and Barrows assays were different, as were the screening formats: arrayed (siRNA) vs. pooled (CRISPR).  And this was one of the earliest CRISPR libraries.  Much has been learned about optimising gRNA efficiency and specificity since the Shalem screen.

However, it is also important to note that consistent inhibition of gene expression does not guarantee consistent phenotypes.  The above analysis suggests that care is needed in interpreting the results of CRISPR screens.  RNAi screens possess advantages, e.g. ease of arrayed screening, that will make them useful for many years to come.

Want to receive regular blog updates? Sign up for our siTOOLs Newsletter:



Follow us or share this post:
The final RNAiL?

The final RNAiL?

A recent article in The Scientist asks whether, in light of a paper by Lin et al. showing phenotypic discrepancies between RNAi and CRISPR, this is not ‘the last nail in the coffin for RNAi as a screening tool’?

The paper in question found that a gene (MELK) that had been shown by many RNAi-based studies to be critical for several cancer types shows no effect when knocked out via CRISPR.  They also report that in relevant published genome-wide screens, MELK was not at the top of the hit lists.

Does this mean that the papers that used RNAi were unlucky and off-target effects were responsible for their observed phenotypes?

Gray et al. identified MELK as a gene of interest based on microarray experiments.  They then designed RNAi experiments to test its role in proliferation.  Assuming that this study and the subsequent ones followed good RNAi experimental design (using reagents with varying seed sequences, testing the correlation between gene knockdown and phenotypic strength, etc.), we can be fairly confident that MELK is involved in proliferation.  It might not be the most essential player, which would explain why it is not at the top of screening hit lists.  And screening lists have the draw-back of enriching for off-target hits.

Another possibility is that Lin et al. have observed a known complicating feature of knock-out screens: genetic compensation.  Although they undertake experiments to address this issue, it could be that compensation takes place too quickly for their experiments to rule it out.  Furthermore, they could have addressed this issue by testing knock-down reagents themselves, and checking whether genes they hypothesise as responsible for the supposed off-target effect in the published RNAi work are in fact down-regulated.  C911 reagents could also be used to test for off-target effects.  This is extra work, but given that they are disputing the results in many published studies, this seems justified.

As regards the role of RNAi in screening, The Scientist concludes with the following (suggesting that their answer to the question of whether this is the final nail is also No):

In the meantime, one obvious solution to the problem of target identification and validation is to use both CRISPR and RNAi to validate a target before it moves into clinical research, rather than relying on a single method. “We have CRISPR and short hairpin reagents for every gene in the human genome,” said Bernards. “So when we see a phenotype with CRISPR, we validate with short hairpin, and the other way around. I think that would be ideal.”

Although we agree that validating CRISPR hits with RNAi reagents is important (especially if drugability is a concern), one has to be careful with RNAi reagents, like single siRNAs/shRNAs or low-complexity pools, that are susceptible to seed-based off-target effects.  For validating CRISPR screening hits, siPOOLs provide the best protection against unwanted off-target effects, saving you time, money, and disappointment during the validation phase.


Want to receive regular blog updates? Sign up for our siTOOLs Newsletter:

Follow us or share this post:
Where’s the beef?

Where’s the beef?

In our last blog entry, we discussed a classic RNAi screening paper from 2005 that showed that the top 3 screening hits were were due to off-target effects.

In this post, we analyse a more recent genome-wide RNAi screen by Hasson et al., looking in more detail at what proportion of top screening hits are due to on- vs. off-target effects.

Hasson et al. used the Silencer Select library, a second-generation siRNA library designed to optimise on-target knock down, and chemically modified to reduce off-target effects.  Each gene is covered by 3 different siRNAs.

To begin the analysis, we ranked the screened siRNAs in descending order of % Parkin translocation, the study’s main readout.

We then performed a hypergeometric test on all genes covered by the ranked siRNAs.  For example, if gene A has three siRNAs that rank 30, 44, and 60, we calculate a p-value for the likelihood of having siRNAs that rank that highly (more details provided at bottom of this post).  It’s the underlying principle of the RSA algorithm, widely used in RNAi screening hit selection.  If the 3 siRNAs for gene B have a ranking of 25, 1000, and 1500, the p-value will be higher (worse) than for gene A.

The same type of hypergeometric testing was done for the siRNA seeds in the ranked list.  For example, if the seed ATCGAA was found in siRNAs having ranks of 11, 300, 4000, and 6000, we would calculate the p-value for those rankings.  Seeds are over-represented in siRNAs at the top of the ranked list will have lower p-values.

After doing these hypergeometric tests, we had a gene p-value and a seed p-value for each row in the ranked list.  We could then look at each row in the ranked list estimate whether the phenotypic is due to an on- or off-target effect by comparing the gene and seed p-values.  [As a cutoff, we said that the effect is due to one of either gene or seed if the difference in p-value is at least two orders of magnitude.  If the difference is less than this, the cause was considered ambiguous.]

After assigning the effect as gene/seed/ambiguous, we then calculated the cumulative percent of hits by effect at each position in the ranked list.   Those fractions were then plotted as a stacked area chart (here, looking at the top 200 siRNAs from the screen):


The on-target effect is sandwiched between the massive ‘bun’ of off-target effects and ambiguous cause.  We are reminded of these classic commercials from the 80s:


Want to receive regular blog updates? Sign up for our siTOOLs Newsletter:


Note on p-value calculations:

P-values were calculated using the cumulative hyper-geometric test (tests the probability of finding that many or more instances of members belonging to the particular group, in our case a particular gene or seed sequence).  The p-value associated with a gene or seed is the best p-value for all the performed tests.  For example, assume a gene had siRNAs with the following ranks: 5, 20, 1000.  The first test calculates the p-value for finding 1 (of the 3) siRNAs when taking a sample of 5 siRNAs.  The next test calculates the p-value for finding 2 (of 3) siRNAs when taking a sample of 20 siRNAs.  And the last is the probability of getting 3 (of 3) siRNAs when taking a sample of 1000.  If the best p-value came from the second test (2 of 3 siRNAs found in a sample size of 20), that is the p-value that the gene receives.  This is also the approach used by the RSA (redundant siRNA activity) algorithm.  One advantage of RSA is that it can compensate for variable knock down efficiency of the siRNAs covering a gene (e.g. if 1 of 3 gives little knockdown).

Follow us or share this post:
Classic Papers Series: Lin et al. show RNAi screen dominated by seed effects

Classic Papers Series: Lin et al. show RNAi screen dominated by seed effects

Over the coming months, we will highlight a number of seminal papers in the RNAi field.

The first such paper is from 2005 by Lin et al. of Abbott Laboratories, who showed that the top hits from their RNAi screen were due to seed-based off-target effects, rather than the intended (and at that time, rather expected) on-target effect.

The authors screened 507 human kinases with 1 siRNA per gene, using a HIF-1 reporter assay to identify genes regulating hypoxia-induced HIF-1 response.

In the validation phase of their screen, they tested new siRNAs for hit genes, but found that they failed to reproduce the observed effect, even when using siRNAs that had a better on-target knock down than the pass 1 siRNAs.

Figure 1A.  Left panel shows on-target knock down of pass 1 siRNA for GRK4 (O) and the new design (N).  Centre panel shows Western blot of protein  levels  Right panel shows HIF-1 reporter activity for positive control (HIF1A) and the original (O) and new (N) siRNAs.

The on-target knock down is much-improved for the new design, yet its reporter activity is indistinguishable from negative control.  Yet the pass 1 siRNA with poor knock down gives almost as strong a result as HIF1A (positive control).

By qPCR, they then showed that GRK4(O) and another one of the top 3 siRNAs silence HIF1A (the positive control gene).  Using a number of different target constructs they also nicely show that it was due to seed-based targeting in the 3′ UTR.

Although the authors screened at a high initial concentration (100 nM), the observed off-targets persisted at 5 nM, suggesting that just screening at lower concentrations would not have improved their results.

The authors conclude:

In addition, due to the large percentage of the off- target hits generated in the screening, using a redundant library without pooling in the primary screen could significantly reduce the efforts required to eliminate off-target false positives and therefore, will be a more efficient design than using a pooled library.

This is true for low-complexity pools, but high-complexity pools can overcome this problem by providing a single reliable result for each screened gene.

Want to receive regular blog updates? Sign up for our siTOOLs Newsletter:

Follow us or share this post:
The limits of chemical modification

The limits of chemical modification

In addition to potential reductions in on-target efficiency, chemically modifying siRNAs will not necessarily eliminate seed-based off-target effects.

Rasmussen et al. found that a chemically-modified siRNA can still have substantial seed effects.

They examined the expression data for 3 siRNAs from Jackson et al. and showed that for one of them the seed is still active following chemical modification.

The algorithm of Rasmussen et al. (cWords) looks through a ranked list of 3′ UTR sequences (in this case, ranked by deregulation as measured on a microarray) and finds words that are enriched towards the top of the list.

The seed target sequence of the unmodified Pik3ca siRNA is strongly enriched (B panel), as expected, but is also strongly enriched after chemical modification (C panel):



Want to receive regular blog updates? Sign up for our siTOOLs Newsletter:

Follow us or share this post:
Deconvoluted SMARTpools are like a box of chocolates

Deconvoluted SMARTpools are like a box of chocolates


Falkenberg et al. (2014) performed a synthetic lethal  RNAi screen to identify genes which, when knocked down in combination with drug treatment, induced apoptosis in drug-resistant cells.

The first screening pass covered over 18,000 protein-coding genes using Dharmacon’s siGENOME, probably the most widely used library for genome-wide RNAi screens.

siGENOME is based on low-complexity pooling, with 4 siRNAs pooled per gene.  In the first pass, hits are identified based on the pooled siRNA result.  In the second pass, siRNAs for hit genes are tested individually (deconvoluted).

In Falkenberg et al.’s pass 2, they examine 450 hit genes, split across 2 phenotypes of interest (Caspase-Glo 3/7 as a measure of apoptosis and Cell Fluor Titre, CTF, as a measure of general viability).  They choose 317 genes for Caspase-Glo 3/7 and 150 genes for CTF (adds to 467 because 17 genes are hits for both phenotypes).

If the pooled result from pass 1 is due to an on-target effect, and if the siRNAs give consistent knockdown, the 4 siRNAs should give similar phenotypes (and they should also be similar to the pooled phenotype from pass 1).

Is that what Falkenberg et al. show?   Or is the pool more like Gump’s proverbial box of chocolates?  The latter looks to be the case, as the following plots show.  For the 2 phenotypes of interest, the top 20 genes (based on the pass 1 pool phenotype) were plotted with the pass 1 (pooled) and pass 2 (single siRNA) results.

Falkenberg_deconvolution_plots 6 Falkenberg_deconvolution_plots 5

With a few exceptions, the siRNA results are widely divergent.

(note that the pass 2 phenotypic results are weaker than pass 1, and the authors discuss the reduction in dynamic range in the second pass– however, the divergence between individual siRNAs is still pronounced)

As suggested by the above plots, the correlation between different siRNAs for the same gene is very weak:

Falkenberg_deconvolution_plots 8Falkenberg_deconvolution_plots 7

Note that R’s of 0.15 and 0.12 mean that only 2.3% and 1.4%, respectively, of the screening variance can be explained by on-target effects of the reagent.

In light of this very weak on-target signal in pass 2, it’s surprising that Dharmacon reported to the authors (after analysing their data) that there were no seed-based off-target effects.  Given these results, and the known preponderance of seed-based off-target signal and the necessity of high-complexity pools to overcome seed effects, that statement must be treated very skeptically.

Overall, it looks like we can be quite confident about the top screening hit, but beyond that it really difficult to know what is causing the observed phenotype (probably not on-target knockdown effects).

To paraphrase the wisdom of Forest Gump, deconvoluted Dharmacon pools are like a box of chocolates: you never know what you’re going to get!


Additional info:

Read More Read More

Follow us or share this post:
Simplicity is the ultimate sophistication

Simplicity is the ultimate sophistication

The beauty of the siPOOL strategy is its simplicity.

In this presentation from the  (relatively) early days of Apple, Steve Jobs says that his company’s goal is to serve the one-on-one relationship between a user and his/her computer.


Similarly, siPOOLs, are designed to serve the one-on-one relationship between a scientist and his/her RNAi results.

By providing an interpretable result without the need for extensive follow-up work and off-target corrections, siPOOLs make it possible for a scientist to use a single gene list to gain insight into biological function.

We believe, as stated in the brochure to market the Apple II,  that  Simplicity is the ultimate sophistication.

(Note: this quote is popularly, though apparently falsely, attributed to Leonardo da Vinci) .


Want to receive regular blog updates? Sign up for our siTOOLs Newsletter:

Follow us or share this post:
Seed effects persist in hyperdimensional space

Seed effects persist in hyperdimensional space

Work from the Carpenter lab suggests that attempts to shake seed-based off-targets by going to  ‘phenotypic hyperspace’ will not work.

They performed a high-content assay with 315 shRNAs covering 41 genes.  A 1301-dimensional profile was created for each well, and compressed to 205 principal components that captured  99% of the variance.

The hope would be that by examining a wider phenotypic space, the gene-specific effects of RNAi reagents would become more prominent.

However, the profiles between shRNAs targeting the same gene are only slightly better than those between random shRNAs, while shRNAs sharing the same seed sequence have much more similar profiles.

Screen Shot 2015-12-15 at 14.27.51

(figure shows percent of significant profile correlations for different pairings)

Off-target phenotypes can only be escaped by using a reagent that exclusively knocks down the target gene.

Want to receive regular blog updates? Sign up for our siTOOLs Newsletter:

Follow us or share this post:

Like what you see? Mouse over icons to Follow / Share