There is no way to "isolate" $r$ but we can approximate it. The equation being
$$A = P\left(1 + \frac rn\right)^{nt} + \frac{PMT\left(\left(1 + \frac rn\right)^{nt} - 1\right)}{\frac rn}$$ let $$a=\frac PA\qquad b=MT \qquad c=nt \qquad x=\frac r n$$ to make it
$$a=\frac{x}{(b+x) (x+1)^c-b}$$ and we know that $x \ll 1$.
Developing the rhs as a Taylor series built at $x=0$, we have
$$a=\alpha+\beta x+\gamma x^2+\delta x^3+O\left(x^4\right)$$
$$\alpha=\frac{1}{b c+1}\qquad \beta=-\frac{c (b(c-1)+2)}{2 (b c+1)^2}\qquad \gamma=\frac{c (c+1) (b (c-1) (b c+4)+6)}{12 (b c+1)^3}$$ $$\delta=-\frac{c (c+1) (b (c-1) (c (b (b c+4)+1)+6)+4 (c+2))}{24 (b c+1)^4} $$Now, using series reversion
$$x=y-\frac{\gamma }{\beta } y^2+\frac{2 \gamma ^2-\beta \delta
}{\beta ^2}y^3 +O\left(y^4\right) \qquad \text{where} \qquad y=\frac{a-\alpha }{\beta }$$
It is easy to get more terms in $y^m$ and have better and better approximations for $x$.