Does the outputClean LDA model generalize to other species?

[davidfstein]

Hi,

Do you anticipate that the trained LDA model in outputClean will work well for non-human species? I am testing it on some mouse probes and receiving strange results. Particularly, for probes that bowtie2 marks as having several alignments, outputClean passes 100 percent of the candidates when I would expect fewer to pass. This is regardless of how I set the p flag (even -p 1 which seems strange). Maybe there are some caveats I need to be aware of when using the model with outputClean. I'm attaching my input fasta file in case that sheds some light.

Thanks!
mouse.txt

[brianbeliveau]

Hi David, Yes, we have used the LDA mode to design probes for multiple species without issue. Looking at your input, I think the issue is that your FASTA header does not match what the scripts expect: it needs to be chrom:start-stop using the same scaffold information as the reference genome, e.g. chr1:100000-100035. This info gets passed to ouputClean via the .fastq header for each probe candidate, and outputClean will not run properly if it cannot find valid values for chrom, start, stop. You should see the issue go away if you either alter your FASTA header or set using the -H flag in blockParse. Best, Brian

…

------------------------------------------------ Brian J. Beliveau, Ph.D. Assistant Professor of Genome Sciences University of Washington He/him/his +1 978-390-1933 | beliveau.io

On Mon, Jun 29, 2020 at 12:33 PM David Stein ***@***.***> wrote: Hi, Do you anticipate that the trained LDA model in outputClean will work well for non-human species? I am testing it on some mouse probes and receiving strange results. Particularly, for probes that bowtie2 marks as having several alignments, outputClean passes 100 percent of the candidates when I would expect fewer to pass. This is regardless of how I set the p flag (even -p 1 which seems strange). Maybe there are some caveats I need to be aware of when using the model with outputClean. I'm attaching my input fasta file in case that sheds some light. Thanks! mouse.txt <https://github.com/beliveau-lab/OligoMiner/files/4847634/mouse.txt> — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#11>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHIXQ2UOMNR6M5K4IPKQ27TRZDT5DANCNFSM4OLQYC5Q> .

[davidfstein]

Hi Brian,

Thanks for your help! Does the model actually use the chrom:start-stop information as features for the classification, or only as a way to parse the sam file? I tried it again with ">Tubb3 chr8:123411553-123422015" but I guess I'm still using the wrong header. Looking at the accession on NCBI, https://www.ncbi.nlm.nih.gov/nuccore/NC_000074.6?report=genbank&from=123411553&to=123422015, I thought that my header was correct, sorry if this is outside the scope, but do you know the best way for me to find the correct "chrom:start-stop" metadata when retrieving new sequences?

Best,
David

[brianbeliveau]

Hi David, The chrom:start-stop is used in the parsing of the sam file, which is important in deciding which candidate probes get passed to the LDA model vs put directly in the output. This is likely what caused the issue you saw. The header you would want is ">chr8:123411553-123422015". I would generally recommend pulling sequence info from UCSC if not from an existing annotation file. Hope that helps, Brian

…

------------------------------------------------ Brian J. Beliveau, Ph.D. Assistant Professor of Genome Sciences University of Washington He/him/his +1 978-390-1933 | beliveau.io

On Mon, Jun 29, 2020 at 1:19 PM David Stein ***@***.***> wrote: Hi Brian, Thanks for your help! Does the model actually use the chrom:start-stop information as features for the classification, or only as a way to parse the sam file? I tried it again with ">Tubb3 chr8:123411553-123422015" but I guess I'm still using the wrong header. Looking at the accession on NCBI, https://www.ncbi.nlm.nih.gov/nuccore/NC_000074.6?report=genbank&from=123411553&to=123422015, I thought that my header was correct, sorry if this is outside the scope, but do you know the best way for me to find the correct "chrom:start-stop" metadata when retrieving new sequences? Best, David — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#11 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHIXQ2VRPHQK4BZBVLK4LELRZDZJHANCNFSM4OLQYC5Q> .

[davidfstein]

Ok, I'm able to get it working, thanks!