Result interpretation for a SOX6 variant

[sms07207]

I have a variant (chr11-16049712-C-A; GRCh38). This variant's MANE select transcript refers to NM_001367873(SOX6):c.1435+43G>T

This is the spliceAI prediciton:

It has a donor loss at 1bp position; however for the MANE select transcipt, it's a +43 intronic variant and that does not have a canonical splice donor site to lost. May I ask if I have any misunderstanding in the interpretation of spliceAI results? Thank you.

[bw2]

I see your point.. when looking at the visualizations for (NM_001367873(SOX6):c.1435+43G>T), the canonical MANE transcript donor site is far away from +1bp.

To me this means the model is predicting that the variant could disrupt any transcripts that share the same transcript start and end as ENST00000683767.1 but use the +1bp donor rather than the +43 donor (if such transcripts exist). At the same time, the SpliceAI score visualizations show almost no effect on the +43 donor used by the MANE transcript, so the model does not expect the MANE transcript to be disrupted.

The current user interface confusingly shows the high score next to the MANE transcript. This only means that the model generated its predictions based on the transcript start and end positions (and strand) of ENST00000683767.1. The model is not aware of exon boundaries when it generates these predictions.

The "mask" behavior for this variant is similarly confusing but correct. SpliceAI post-processes the predicted scores to zero out any scores that strengthen annotated splice sites or weaken unannotated splice sites. For the ENST00000683767.1 transcript, the +1 position is not an exon boundary and so SpliceAI treats it as "unannotated" and therefore masks it.

As a side note, I've double-checked that SpliceAI-lookup is running SpliceAI with the correct transcript and exon boundaries in this case.