Simple pytorch/pyro implementation of the hill equation in a bayesian regression framework. Allows for placement of priors on variables as well as propagation of uncertainty onto the summary metrics like IC50.
We use the model: $$ y = E_0 + (\frac{ E_{max} - E_0 } {1 + (\frac{EC_{50}}{x})^H}) $$
Where,
y : cell viability [0,1]
x : log10 concentration
$E_0$ : minimum inhibition (max cell viab.)
$E_{max}$ : maximum inhibition (min cell viab.)
$EC_{50}$ : concetration at which there is 50% maximum inhibition (e.g., conc (x) at which y = ($E_0$ -$E_{max}$ )/2 +$E_{max}$ )
H : Hill coefficient
We believe this method is especially valuable in low data scenarios, or when there is poor concordance between replicates.
See the tutorial for example on how to use this code.
Email evansna@ohsu.edu for questions.