ShenLab/MisFit

Estimation of selection coefficient of missense variants using human population genomes and machine learning

MisFit

predicting nonsynonymous variant selection coefficient

population genetics model

pop_model simulate variants and construct PIG model

protein-truncating variants

model_PTV only use PTVs, independent of other models

prior of missense variants

model_mis used to find priors of d and s_gene, then initialize s_gene before MisFit training.

Baseline models

model_basic population data w./w.o. genes

model_logit population data + gene + ESM zero-shot as d

MisFit model

model_TF full MisFit model

*_analysis are used to combine data for different analysis

Note: model_selection directly given by the model may need to be transformed by a sigmoid function to get MisFit_S in the original scale

evaluation and figure-plotting

model_evaluate

data processing

to be updated

• deep mutational scan GMM
• variant annotations