What is CI-SpliceAI and how is it different from SpliceAI?
Learn what CI-SpliceAI is, and how it is integrated into SeqOne
CI-SpliceAI (Y Strauch et al., 2022) is based on the SpliceAI algorithm and relies on its own neural network model based on validated transcripts from the Gencode cohort.
Description
CI-SpliceAI is a splice deep neural network that was trained to detect splicing sites. This prediction score is based on the same engine that was developed in SpliceAI but the model was re-trained with manually annotated splice sites from GENCODE and without level 3 transcripts (automated annotation) to only use manually annotated transcripts. CI-SpliceAI was trained on GENCODE annotations (v37GRCh38), and on all chromosomes (unlike SpliceAI).
CI-SpliceAI authors show that using this manually curated set of transcripts enables a higher level of accuracy in the predictions. Additionally, CI-SpliceAI is able to annotate all variants including MNVs.
This score is very similar to SpliceAI but offers greater accuracy. It provides prediction scores to identify variants with an impact on alternative splicing. On SeqOne, the scores are available for variants meeting the following criteria:
- noted by VEP: splice_*
- Freq. GnomAD < 10⁻³
- Quality (QA > 20, AO > 10, DP > 20)
Interpretation
The column gives 4 delta scores: acceptor site loss or gain (AL, acceptor loss and AG, acceptor gain) and donor site loss or gain (DL, donor loss and DG, donor gain). Delta scores range from 0 to 1 and can be interpreted as the probability that the variant affects splicing at any position within a window around it (+/- 500bp by default). In the SpliceAI paper, a detailed characterization is provided for 0.2 (high recall), 0.5 (recommended), and 0.8 (high precision) cutoffs. Consequently, we defined the same 3 levels of scores:
- High recall (score > 0.2): appearing in green in the variant viewer
- Recommended (score > 0.5): yellow
- High precision (score > 0.8): red
SpliceAI
For each variant, the SeqOne interface provides a link to the SpliceAI lookup site where users can check the SpliceAI score.