GRCh38 model

[gianfilippo]

Hi,

do you have a pretrained model based on GRCh38 ?

Thanks Gianfilippo

[suragnair]

Unfortunately not

[gianfilippo]

thanks Gianfilippo

On Sat, Jul 17, 2021 at 8:36 PM Surag Nair @.***> wrote:

Unfortunately not

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/kundajelab/ChromDragoNN/issues/9#issuecomment-881941090, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACPSFVHMBZWSYNCZ7YQJE33TYHESLANCNFSM5APY5ELQ .

[akundaje]

Note that you can use the model on any genome version since it only needs a sequence and regulator expression input. The sequence can be from any human genome version

On Sat, Jul 17, 2021, 12:48 PM Gianfilippo Coppola @.***> wrote:

thanks Gianfilippo

On Sat, Jul 17, 2021 at 8:36 PM Surag Nair @.***> wrote:

Unfortunately not

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub < https://github.com/kundajelab/ChromDragoNN/issues/9#issuecomment-881941090 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/ACPSFVHMBZWSYNCZ7YQJE33TYHESLANCNFSM5APY5ELQ

.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kundajelab/ChromDragoNN/issues/9#issuecomment-881951067, or unsubscribe https://github.com/notifications/unsubscribe-auth/AABDWEIJBQVWJNIMA3BLI4DTYHNCHANCNFSM5APY5ELQ .

[gianfilippo]

Hi,

yes, I understand I can train the model on GRCh38 based data. I was hoping you already had a GRCh38 trained model I could use for prediction.

Perhaps, if the gene definition for the predictors you use in your model is the same,, I could use the expression levels I have as input and get hg19 based DHS levels from the available model, then, convert the results to GRCh38?

I may have already asked this, but how do you handle batch effects ?

Thanks Gianfilippo

On Sat, Jul 17, 2021 at 10:04 PM Anshul Kundaje @.***> wrote:

Note that you can use the model on any genome version since it only needs a sequence and regulator expression input. The sequence can be from any human genome version

On Sat, Jul 17, 2021, 12:48 PM Gianfilippo Coppola @.***> wrote:

thanks Gianfilippo

On Sat, Jul 17, 2021 at 8:36 PM Surag Nair @.***> wrote:

Unfortunately not

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <

https://github.com/kundajelab/ChromDragoNN/issues/9#issuecomment-881941090

, or unsubscribe <

https://github.com/notifications/unsubscribe-auth/ACPSFVHMBZWSYNCZ7YQJE33TYHESLANCNFSM5APY5ELQ

.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub < https://github.com/kundajelab/ChromDragoNN/issues/9#issuecomment-881951067 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AABDWEIJBQVWJNIMA3BLI4DTYHNCHANCNFSM5APY5ELQ

.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/kundajelab/ChromDragoNN/issues/9#issuecomment-881953078, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACPSFVCCME3QQSVHE2OOAOLTYHO3TANCNFSM5APY5ELQ .

[akundaje]

My point is training on GRCh38 or GrCh37 doesn't matter. A trained model can be used for prediction on any sequence (from any human genome version or synthetic sequences) and cell type with available expression data. So if you want to make predictions you can use the trained model as is.

We don't handle batch effects or data quality differences in this version. For now, it would be to be taken care of in the data preprocessing.

We will be explicitly modeling these effects in the next iterations of this work.

Anshul

On Sun, Jul 18, 2021, 7:45 AM Gianfilippo Coppola @.***> wrote:

Hi,

yes, I understand I can train the model on GRCh38 based data. I was hoping you already had a GRCh38 trained model I could use for prediction.

Perhaps, if the gene definition for the predictors you use in your model is the same,, I could use the expression levels I have as input and get hg19 based DHS levels from the available model, then, convert the results to GRCh38?

I may have already asked this, but how do you handle batch effects ?

Thanks Gianfilippo

On Sat, Jul 17, 2021 at 10:04 PM Anshul Kundaje @.***> wrote:

Note that you can use the model on any genome version since it only needs a sequence and regulator expression input. The sequence can be from any human genome version

On Sat, Jul 17, 2021, 12:48 PM Gianfilippo Coppola @.***> wrote:

thanks Gianfilippo

On Sat, Jul 17, 2021 at 8:36 PM Surag Nair @.***> wrote:

Unfortunately not

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <

https://github.com/kundajelab/ChromDragoNN/issues/9#issuecomment-881941090

, or unsubscribe <

https://github.com/notifications/unsubscribe-auth/ACPSFVHMBZWSYNCZ7YQJE33TYHESLANCNFSM5APY5ELQ

.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <

https://github.com/kundajelab/ChromDragoNN/issues/9#issuecomment-881951067

, or unsubscribe <

https://github.com/notifications/unsubscribe-auth/AABDWEIJBQVWJNIMA3BLI4DTYHNCHANCNFSM5APY5ELQ

.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub < https://github.com/kundajelab/ChromDragoNN/issues/9#issuecomment-881953078 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/ACPSFVCCME3QQSVHE2OOAOLTYHO3TANCNFSM5APY5ELQ

.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/kundajelab/ChromDragoNN/issues/9#issuecomment-882067879, or unsubscribe https://github.com/notifications/unsubscribe-auth/AABDWENFBJEE2CNYJME7A3TTYLSJDANCNFSM5APY5ELQ .