SpliceMap comes packaged with a number of filters to help you control the specificity of the outputted junctions. Feel free to mix and match the filters on your data. These filters only work on the junctions (.bed files). In future we may implement some filters for the SAM file. These filters are perhaps more useful for longer read length as a greater total number of junctions are found.
Click on the name of the filter for usage instructions. These filters are also described in detail in the SpliceMap paper.
nnrFilter is the strictest filter. It filters reads based on the number of non-redundant supporting reads. Non-redundant is defined as a read that is not mapped to exactly the same chromosome position. Empirical results show that using nNR >= 2, will provide a specificity of about 90%.
uniqueJunctionFilter removes junctions which are solely supported by multiply mapped reads. That is, junctions which have at least one uniquely mapped supporting read. Only junctions with nNR = 1 are targeted by this filter.
Neighbor filter (BETA)
neighborFilter removes "lonely" junctions. Lonely is defined as a junction with no exonic reads nearby. This filter seems to be able to remove the few artifact junctions that appear in the middle of no where. However, you may not want to use this filter if you are looking for low coverage junctions. Only junctions with nNR = 1 are targeted by this filter.
The following table summarizes the effect of the various filters on the sensitivity and specificity. Results are from 20 million 100bp RNA-seq reads mapped to the human genome (hg18). EST validation was used to as a judge of specificity.
|Unfiltered||nUM||Neighbor (160bp)||nUM+Neighbor (160bp)||nNR|