Question on parameters

[sophiachen1]

Hi @kmnip ,

Thank you so much for creating RNA-Bloom! I have a quick question about the parameters.
What does a "tip" mean and what does the parameter "-tiplength" control?
The "-overlap" parameter is "minimum number of overlapping bases between reads [k-1]", is it recommended to use kmer size-1? and how do you choose 150 for ONT cDNa seq data?
If I can tolerate false positives in the transcripts to some extent, beside adjusting fpr to a higher value like 0.3, do you suggest to add other options, like -sensitive, -artifact or -chimera?

Thanks for your help!
Sophia

[kmnip]

Hi @sophiachen1 ,

Thanks for your interest in using RNA-Bloom!

I recommend using the default parameters and only adjust them if you find issues in your assembly:

java -jar RNA-Bloom.jar -long LONG.fastq -t THREADS -outdir OUTDIR

What does a "tip" mean and what does the parameter "-tiplength" control?

A tip is a short dead-end branch in the graph. In long-read assembly, -tiplength controls how the overlap between reads are treated. The default is 50 when -long is applied.

As an example, you have two overlapping reads, and the TTT in reads 1 and 2 are tips of length 3:

Read 1:          TTT====AAAAAAAAAA
Overlap:            ||||
Read 2:  GGGGGGGGGGG====TTT

For -tiplength 4, tips of length 4 or less are tolerated, and the example above is treated as a true overlap between reads 1 and 2. These two reads can be assembled into:

GGGGGGGGGGG====AAAAAAAAAA

If -tiplength 0, then the overlap between reads 1 and 2 is discarded.

This parameter is intended to handle erroneous basecalls near end of ONT reads. It is highly recommended to trim adapters before assembly.

The "-overlap" parameter is "minimum number of overlapping bases between reads [k-1]", is it recommended to use kmer size-1? and how do you choose 150 for ONT cDNa seq data?

The default of k-1 in the help menu was intended for short read assembly. When -long is applied, this is defaulted to 150 as you have noted. This was determined based on benchmarking on simulated and experimental ONT data from several years ago. It balances assembly length and misassemblies. Increasing -overlap would reduce misassemblies, but it may not perform as well for low-expressed transcripts, due to the lower likelihood of long overlaps between small number of reads.

If I can tolerate false positives in the transcripts to some extent, beside adjusting fpr to a higher value like 0.3, do you suggest to add other options, like -sensitive, -artifact or -chimera?

-fpr is intended to set the false positive rate of the Bloom filters used during assembly. This is not same as the FPR of the output assembly. Increasing it will reduce the memory usage of Bloom filters, whereas lowering it will increase the memory usage. If you increased this to 0.3 (i.e. 30%), then on average 3 in 10 k-mers in the Bloom filters would be false positives!

The other parameters -sensitive, -artifact or -chimera are intended for short-read assemblies, and you can safely ignore them.

Hope that helps!
Ka Ming

[sophiachen1]

Thank you @kmnip for your quick and clear response!

[sophiachen1]

Hi @kmnip , I am also wondering if there's quantification information in the output, like how many reads support the transcripts? Thanks!

[kmnip]

@sophiachen1 There is an output field c for coverage, but it is highly inaccurate. I wouldn't relying on it for any type of analyses. You would get more mileage by aligning the reads yourself.