Thread by @thatdnaguy

#ASHG20 BH
Inference of fitness effects of short tandem repeat polymorphisms improves functional categorization.
Bonnie Huang.

#ASHG20 BH
@b_b_huang
Looking at short tandem repeats (STRs). 1-6 bp repeat unit. 1.6 M loci in human genome. Implicated in >40 disorders.

#ASHG20 BH STRs are highly polymorphic. Hard to distinguish from benign and de novo STR mutations. Harder to interpret than SNVs in general.

#ASHG20 BH They are interested in STR mutation. Lack of tools to score pathogenicity. Developed SISTR method to score the potential fitness effects in coding and non-coding STRs.
bit.ly/3mAaTY3

#ASHG20 BH At each locus outputs a selection coefficient (S) that can be interpreted as a pathogenicity score. Higher scores more likely to be pathogenic.

#ASHG20 BH How to infer S? Generalized step-wise mutation model with two mods. 1) has directional bias in mutation sizes (longer more likely to contract. shorter more likely to expand). 2) Length-dependent mutation rate (longer more likely to mutate).

#ASHG20 BH Fitness model. Assumption that the model or central allele at each STR has optimal fitness. Inference is approximate Bayesian model. Forward simulate 50k generations. Compared simulated frequency to observed. If the allele frequencies are similar, accept value of S.

#ASHG20 BH Validated it using simulated data with various mutation models. SISTR recovers simulated values down to S=10^-4 for most settings. Limitation - for certain classes of loci hard to measure selection coefficients, esp loci with low mutation rate.

#ASHG20 BH Can you prioritize de novo mutations? Look at autism spectrum disorder (ASD) quads (mom, dad, unaffected sib, affected sib). Wanted to know if mutations with high SISTR scores more likely to be pathogenic.

#ASHG20 BH Top candidate pathogenic de novo STR - AP4S1, PDCD1, CEP120, MED13L, KCNB1, AGO1, CACNA2D3, RFX3, SAIH3, LINC00534, GLIS3, FOXP1.

#ASHG20 BH Related preprint to SISTR in ASD:
bit.ly/31Y3gTh