Illumina QC FAQs

Expand all Collapse all

• What's a Q score?

  Illumina uses Phred quality scores. A Phred quality score is produced by a model that uses quality predictors as inputs and produces Q scores as outputs. Illumina's algorithm uses many predictors to generate a Q score for each base in every read.. These predictors can be divided into three categories: single base, full-read and “neighborhood”.

  A Q score is simply a representation of the probability that a base-call is incorrect. Here's the equation: Q = -10 * log₁₀p where p is the probability that the particular base-call is incorrect (calculated using the set of predictors). A base-call with a Q score of 20 has a 1% probability of being incorrect; Q30 corresponds to 0.1%, Q40 to 0.01%, and so on.

  Note about offsets:

  Within your sequencing files Q scores are generally represented using a single ASCII character. Each ASCII character has a value that corresponds to the Q score value based on some offset. Sanger Q scores use 33 as the offset value. Currently Illumina also uses 33, but that wasn't always the case. Older Illumina data used 64 as an offset. This isn't really a big deal. The quality of the data isn't affected, but some secondary analysis tools will need to be told which offset has been used. You can find this information in the header of the Per Base Sequence Quality graph.

• What is FastQC?

  FastQC is a quality control tool for next-generation sequencing data developed and maintained by a group at the Babraham Institute in the UK. FastQC produces graphs and reports about your next-gen data.

• What is FastQC not?

  FastQC is not the last and final word about the quality of your library or sequencing run. The reports rely on the assumption that all sequencing data comes from completely random, completely diverse starting material. This is not always a good assumption.

• How should the results be interpreted?

  It is important to realize that the FastQC is expecting to see diverse, random sequencing data. You must think about whether this assumption is valid for your organism or library type. You should treat the summary evaluations as pointers to where you should concentrate your attention and understand why your library looks the way it does. Look closely at the graphs to see if they match your expectations.

  For paired-end runs you will receive two FastQC reports per library. One for read one, and one for read two.

  Please be aware that sequencing center staff look at these reports too. If we see something significant we will let you know.

• Per Base Sequence Quality

  The chart:

  X-Axis – Cycles

  Y-Axis – Q score

  There is a whisker box for each cycle. The red line is the median. The yellow box is the inter-quartile range (25-75%). The upper and lower whiskers are at 10% and 90%. The blue line is the mean.

  The Per Base Sequence Quality report does a great job of showing you what happened during the sequencing run. The first few bases will generally be in the low to mid 30s. The Q score should stabilize in the upper 30s and gradually decline over the last half of the run. A good 100 cycle run will look a lot like this:

  

• Per Sequence Quality Scores

  The chart:

  X-Axis – Mean read Q score

  Y-Axis – Reads

  The Per Sequence Quality Score shows you the quality distribution of your library. A mean Q score is calculated for each read, the count for each Q score is plotted on the graph.

  Example from a very good run:

  

• Per Base Sequence Content

  The chart:

  X-Axis – Cycles

  Y-Axis – Percentage of total bases

  The Per Base Sequence Content displays the percentage of each base at each cycle of the sequencing run. This report must be compared with what you expect based on your knowledge of the organism.

  It is very common to see bias at the 5' end of many libraries. Here are two examples from successful runs.

   

   

  

• Per Base GC Content

  The chart:

  X-Axis – Cycles

  Y-Axis – Percentage of total bases

  The Per Base GC Content displays the per cycle GC percentage. This report must be compared with what you expect based on your knowledge of the organism.

  It is very common to see bias at the 5' end of many libraries. Here are two examples from successful runs.

   

   

   

  

• Per Sequence GC Content

  The chart:

  X-Axis – Mean read GC content

  Y-Axis – Number of reads

  The Per Sequence GC Content displays the distribution of GC content. The blue curve represents the theoretical distribution, which is based on your library's GC composition. The red curve is the distribution calculated from your library.

  Examples:

  

  

  

• Per Base N Content

  The chart:

  X-Axis – Cycles

  Y-Axis – Percentage of N's

  The red line shows the percentage of N's at each cycle.

  The HiSeq2000 produces very few Ns. It is very rare to see N content greater than 30%. When Ns are produced it is usually the result of some temporary instrument issue. For example a small bubble in the flow cell may cause focus problems at a certain cycle.

  Downstream processing of Ns depends on your analysis software and strategy.

   

   

  No N's:

  

   

  N spike at cycle 45, but still very good:

  

• Sequence Length Distribution

  The chart:

  X-Axis – Read Length

  Y-Axis – Number of reads

  The red line shows the length of each read

  Since the number of cycles in an Illumina run is fixed, this report will always be an inverted V. There's not much to see here.

  Example from 51 cycle run:

  

• Sequence Duplication Level

  The chart:

  X-Axis – Sequence duplication level

  Y-Axis – Percent duplicate vs unique

  This graph displays a measurement for sequencing duplication using data from the first 50 cycles from a subset of the total reads. The 10 column includes all duplication levels greater than or equal to 10.

  This graph is more challenging than the others. Start by looking at the sequence duplication level. A low duplication level indicates a high percentage of unique reads. A high duplication level indicates a high percentage of duplicate reads. The duplication level of 1 is set to 100. All of the other percentages are calculated based on the number of reads with a certain duplication level relative to the number of unique reads.

  The two most likely causes of duplicate reads are having “too many” reads for a relatively small target and enrichment bias. If you have many millions of reads and a small target you should expect this report to show a high sequence duplication level.

  It's generally a good idea to align your data to check for complete and even coverage.

   

  Example of a small genome size (“over-sequencing”):

   

  Example from a larger target:

  

• Overrepresented Sequences

  A list of sequences that comprise more than 0.1% of the total sequence data.

  This report uses data from the first 50 cycles from a subset of the total reads.

  Overrepresented sequences include the sequences that are highly duplicated in your library, as well as any primer and/or adapter dimers that were present in the original library. Adapter sequences are always present in a sequencing experiment at some level, but aren't problematic in small percentages. These adapters will not align to your genome. They can be ignored, or you may use analysis software to remove them.

• Kmer Content

  The chart:

  X-Axis – Cycle

  Y-Axis – Relative enrichment

  The chart shows up to six 5mers that are overrepresented. The table lists all 5mers that are enriched more than 3-fold overall, or more than 5 fold at any individual cycle. This will show if you have a general enrichment, or if there is a pattern of bias at different points over your read length.

  This module counts the enrichment of every 5-mer within the sequence library. It calculates an expected level at which this k-mer should have been seen based on the base content of the library as a whole and then uses the actual count to calculate an observed/expected ratio for that k-mer. Overrepresented K-mers are a mix of true repeat regions in the genome, and some low quality regions of sequence.