Comparing raw and log-transformed regression coefficients

Question

I'm attempting to perform a power-calculation for a two-sample Mendelian randomization study using genome-wide association study (GWAS) summary statistics, with a binary outcome [http://cnsgenomics.com/shiny/mRnd/]. Observational evidence suggests that the relative risk of my outcome increases by 7% for every 1 unit/day increase in my exposure variable (~Odds Ratio 1.07 per unit/day increase). However, the beta coefficients for the genetic exposures are provided on the log10(units/day) or log10(units/week) scale. Is there a valid method to harmonize the observational and experimental estimates so that they're both on the same scale? This question seems like it might be relevant.