SpliceAI

[damiansm]

https://github.com/Illumina/SpliceAI

[damiansm]

After discussions probably best strategy is to give all spliceAI predicted variants with a maxScore(acceptorGain, acceptorLoss, donorGain, donorLoss) above a threshold (0.2 for high recall and likely pathogenic vars, 0.5 recommended setting for pathogenic vars, 0.8 for high precision pathogenic vars) a high pathogenicityScore (if consider fully LoF should be 1).

Assessing whether these variants are pathogenic really relies on the genomic context of the predicted splice acceptor or donor and size of the effect and can only be confirmed with functional assays.

We can possibly assess performance on a set of TPs from retinal cases and then select the best combination of:

• use maxScore of different methods for acceptor and donor gains and losses
• score threshold to use
• pathogenicityScore ot assign

We also need to implement the best way of obtaining the spliceAI scores: lookup from a pre-calculated table or dynamically calculate

[pnrobinson]

@damiansm can we compare this to Daniel's score?

[damiansm]

@pnrobinson absolutey! that is the plan. I am sure like most predictive methods that using multiple lines of evidence will be the best in the end.

[ShrutiMarwaha]

Hi Damian,

I was wondering if the current (or future version) of exomiser uses SpliceAI? If the current version does, what parameter should I be using to include it in my yaml file?

Thanks, Shruti

[damiansm]

Hi Shruti,

Not currently integrated. We definitely have plans to incorporate this and other splicing approaches in the future but no timeline yet

Cheers Damian

On Thu, May 28, 2020 at 12:01 AM Shruti Marwaha [email protected] wrote:

Hi Damian,

I was wondering if current (or future version) of exomiser uses SpliceAI? If current version does, what parameter should I be using to include it in my yaml file?

Thanks, Shruti

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/exomiser/Exomiser/issues/308#issuecomment-634988361, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABHO4PAZNIW7WIT6JIWESNDRTWLU7ANCNFSM4GRRQ6KQ .

[visze]

Hi,

CADD 1.6 integrates now spliceAI and MMSplice. See the release notes: https://cadd.gs.washington.edu/static/ReleaseNotes_CADD_v1.6.pdf

Of course a general score is not able to beat specific scores in predicting splice-predictions but the new CADD improves splice-predictions a lot.

[ShrutiMarwaha]

Thanks Damian.

[damiansm]

I have successfully generated a tabix score files for SpliceAI SNVs that can be run as a local source. Look to more tightly incorporate scores into the database as a selectable source