No matched normal

Question

No matched normal

crutching opened this issue 7 years ago · 6 comments

Hi, and thanks for all the hard work on this. We are attempting to use your tool, but are working with smaller targeted panel data, without a matched normal. The approach we are taking involves creating a pooled normal sample out of 20 random "normal" samples we have available. I then down sample that file to make the BAM roughly the same size as the tumor samples we are working with. The result is elevated MSI scores across the board (30%-70% range). Most of the samples tend to be in the 40-45% range using your program, so I'm thinking with additional normal samples to add to our pool, we may be able to set MSI-H/L/S cutoffs. Could you share any of your expertise as to whether you think this is plausible? What issues do you think we might encounter?

Answer 1 · 2018-06-10T09:13:41.000Z

We are working on a solution to compute MSI on single tumor sample. Can you share more information about normal panel? We can work out a method together.

…

Hi, and thanks for all the hard work on this. We are attempting to use your tool, but are working with smaller targeted panel data, without a matched normal. The approach we are taking involves creating a pooled normal sample out of 20 random "normal" samples we have available. I then down sample that file to make the BAM roughly the same size as the tumor samples we are working with. The result is elevated MSI scores across the board (30%-70% range). Most of the samples tend to be in the 40-45% range using your program, so I'm thinking with additional normal samples to add to our pool, we may be able to set MSI-H/L/S cutoffs. Could you share any of your expertise as to whether you think this is plausible? What issues do you think we might encounter? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

Answer 2 · 2018-06-10T23:19:26.000Z

Ok, that's great to hear!

We do amplicon sequencing on about 8500 targets, though that number will be increasing to around 12-15k soon. The targets are almost exclusively coding regions. The normal blood samples are all sequenced in precisely the same way, and we currently have 20 of them. We rarely can actually get matched normals, so we are stuck with a non-matched approach.

I was having trouble getting distinguishable results using 20 full size normal datasets combined in to one BAM file. But, downsampling each to about 5% and then combining them seemed to be a step in the right direction.

Are there certain microsatellite markers that are more stable than others, meaning we could compare our tumor sequence to essentially reference genome microsats? I know in work I have done in the past, you typically would see microsats with differing number of repeat units in different samples, but many times a certain number of repeat units was much more common, and thus perhaps we could compare against it.

Answer 3 · 2018-06-25T10:10:14.000Z

well, now you can try msisensor v0.3 for tumor only data.

Answer 4 · 2018-06-26T16:47:25.000Z

@Beifang Great, thanks for adding this! I'm looking forward to testing this, and will report back with results. Question for you, it seems a normal BAM is still required input? What do you expect this normal BAM will include?

Answer 5 · 2018-06-26T17:38:00.000Z

@Beifang Nevermind about my previous question. I had not properly set up my environment to hit v0.3 instead of v0.2.

Answer 6 · 2019-07-09T20:13:31.000Z

in tumor only mode, could you please share your experience on FFPE samples? I ran msisensor on 31 WES FFPE samples and they all have score >10, so am wondering what cutoffs you use/suggests for the FFPE samples.

Thanks,