2023-04-12
Presentation by Anjali Das
She is using functional annotation to improve finding rare variants which are associated with Alzheimer’s disease.
Excerpts from the talk.
Two types of test:
- Gene based RV association test.
- Deep learning burden test
Types of rare variant association tests:
- burden tests collase rare variants into gene scores. e.g. ARIEL, CAST, WSS Loses power when variant effects are in opposing directions, or many non-causal variants.
- adaptive burden tests, use weights and thresholds. e.g. ADA, aSum, VT
- dispersion tests, test variance of genetic effect, e.g. SKAT, C-alpha, SSU
- combined tests, burden + dispersion tests combined, e.g. SKAT-O, Fisher, MIST
- annotation-based tests, add annotations to existing methods, e.g. STAAR, RAVA-FIRST, FST
Funtional score test (FST) workflow:
- Coding variants, within start/end of genes defined by ensemble
- NC variants, Activity-by-contact (ABC) model predicts promoter and enhancer regions of genes.
- Rare variants only – gnomAD AF < 0.05
- Map to annotations
- Map to ADSP individuals
Things I did not understand:
- Why is the Mahattan plot skewed?
- How does the p-value distribution looks in null data?
Interesting comments:
- Jiayu mentioned about the ESM annotation from this paper: Language models enable zero-shot prediction of the effects of mutations on protein function by Meier, Rao, Verkuil, Liu, Sercu and Rives.