search

XiaoTaoWang / EagleC

A deep-learning framework for predicting a full range of structural variations from bulk and single-cell contact maps
Other
51 stars 8 forks source link

pretrained models with low resolution #40

Open asaki1986 opened 1 month ago

asaki1986 commented 1 month ago

Hi all,

EagleC had offered structural variants with great precision.

We did a high-depth sequencing of several well-known cell lines to test our hi-c experiments modified from classic in-situ HI-C protocol.

We downsampling the hi-c data to 30M pairs (~9G), to test the performance of eaglec for shallow hic data.

We found that variant involving chr7-chr14 of Lncap cells was not called, while hic_breakfinder reported this variants within 100kb resolution, which was listed as follow.

#chr1 x1 x2 chr2 y1 y2 strand1 strand2 resolution -logP chr7 12400000 13900000 chr14 36400000 37500000 + + 100kb 74.1075

I am wondering whether eaglec can provided some pretrained models with low resolutions, for example 100k, 1Mb or tutorial to train these models by our own, since the downloaded pretrained models had three resolution from 5k to 50k.

Best, Junfeng

XiaoTaoWang commented 3 weeks ago

Hi Junfeng, the command ‘predictSV-single-resolution’ supports predicting SVs at the 500kb resolution, which might satisfy your needs.

asaki1986 commented 1 week ago

Hi Junfeng, the command ‘predictSV-single-resolution’ supports predicting SVs at the 500kb resolution, which might satisfy your needs.

Thanks,

predictSV-single-resolution successfully detect the variants under 500kb.

I am wondering if eagleC can support SV calling under 1k 100k or 1M resolution, just as selecting variant with eaglec explorer?

Junfeng

XiaoTaoWang commented 4 days ago

Hi Junfeng, currently predictSV-single-resolution is able to detect SVs at resolutions equal to or greater than 50kb (<=50kb) and at 500kb. However, I am planning to release a new version of EagleC soon, which will be applicable to any resolutions, with notably improved performance. I will keep you posted once the new version is released.

asaki1986 commented 2 days ago

Hi Junfeng, currently predictSV-single-resolution is able to detect SVs at resolutions equal to or greater than 50kb (<=50kb) and at 500kb. However, I am planning to release a new version of EagleC soon, which will be applicable to any resolutions, with notably improved performance. I will keep you posted once the new version is released.

Great, will test once released.