Algorithm 1.

Measuring FDR control with synthetic target genes

Input:

- TF expression $X$ (matrix, $N$observations by $G$ genes)

- $x_n$, $x_g$, and $x_{n,g}$ will denote expression for a given sample $n$ and/or gene $g$.

- A desired FDR level $\alpha$

- The number of simulations $J$

- Sets of indices $S_1,S_2,\dots S_J$ indicating the true regulators in each simulation. We select the number of regulators as max(M, 1), where M is Poisson with mean 2. Then we select regulators randomly without replacement.

- Dose-response curves $f_1,f_2,\dots f_J$ dictating the true response to the regulator in each simulation. We set $f_j\left(x_n\right)=1$ if $x_{n,g}>mean\left(x_g\right)$ for all $g$ in $S_j$ and $f_j\left(x_n\right)=0$ otherwise.

Procedure:

- For $j=1\dots J$:

- For $n=1\dots N$:

- Generate $Y_{n,j}\leftarrow f_j\left(x_n\right)$.

- Run the knockoff filter to obtain pairs $(i,j)$ at FDR $\alpha$.

- For $j=1\dots .J$:

- If $i_j$ is in $S_j$, count $(i,j)$ as correct.

- If $i_j$ is not in $S_j$, count $(i,j)$ as a false discovery.