If you assume that
Y = b0 + b1*x1 + b2*x2 + ... + bn*xn + epsilon
then the mean model is just
Y = b0 + b1*x1 + b2*x2 + ... + bn*xn
Then
dY/dx1 = b1
dY/dx2 = b2
...
dY/dxn = bn
In other words, the regression coefficients ARE the partial derivatives of Y w.r.t. the independent variables.