Algorithm 1 Grouping predicted ligandable points into binding sites.

1: function binding_sites($y_p,points$)

2:     $a\leftarrow 0.3$           ▹ Real values a and b define the interval in which we look for the best c value.

3:     $b\leftarrow 1$

4:     $c\leftarrow 0.5$    ▹ Real value c is the threshold for classification of ligandability predictions in each iteration of the while loop.

5:     $ligandablePoints\leftarrow$ select_ligandable_points($y_p,points,c$)

6:     $bindingSites\leftarrow emptyList$

7:     $iteration\leftarrow 0$

8:     $condition\leftarrow “\mathrm{continue}”$

9:     while $condition$ is not “break” do

10:         $clusters\leftarrow$ components($ligandablePoints$)

11:         for $cluster\in clusters$ do

12:              if $cluster$ properties fit into quantile bounds then

13:                  append $cluster$ to $bindingSites$

14:            end if

15:         end for

16:         $points\leftarrow points-bindingSites.points$

17:         if $condition$ is “increase a” then

18:              $a\leftarrow 0.2·(b-a)+a$

19:         else if $condition$ is “decrease b” then

20:              $b\leftarrow 0.2·(a-b)+b$

21:         end if

22:         $c\leftarrow (a+b)/2$

23:         $iteration\leftarrow iteration+1$

24:         $ligandablePoints\leftarrow$ select_ligandable_points($y_p,points,c$)

25:         $condition\leftarrow$ compute_condition($clusters,iteration,\#points$)

26:     end while

27:     return $bindingSites$

28: end function

29: function compute_condition($clusters,iteration,\#points$)

30:      if $iteration>20$ then return “break”

31:      else if $\frac{\mathrm{len}(clusters.points)}{\#points}>0.2$ then return “increase a”

32:      else if $clusters$ is an $emptyList$ then return “decrease b”

33:      else if $max\left(\mathrm{size}\left(clusters\right)\right)<q_v\left(0.1\right)$ then return “decrease b”

34:      else if $max\left(\mathrm{size}\left(clusters\right)\right)>q_v\left(0.9\right)$ then return “increase a”

35:      else if $max\left(\mathrm{avg\_distance}\left(clusters\right)\right)<q_a\left(0.1\right)$ then return “decrease b”

36:      else if $max\left(\mathrm{avg\_distance}\left(clusters\right)\right)<q_a\left(0.9\right)$ then return “increase a”

37:      else if $max\left(\mathrm{std\_distance}\left(clusters\right)\right)<q_s\left(0.1\right)$ then return “decrease b”

38:      else if $max\left(\mathrm{std\_distance}\left(clusters\right)\right)<q_s\left(0.9\right)$ then return “increase a”

39:      else

40:           return “decrease b”

41:     end if

42: end function

43: function components($points$)

44:     $V\leftarrow points$

45:     $E\leftarrow \left\{(x,y)\right|d_2(x,y)<4\}$            ▹ Two points are connected if they are closer than 4Å

46:     $G\leftarrow (V,E)$                                                       ▹ Create a graph with vertices V and edges E

47:     return connected_components(G)

48: end function

49: function select_ligandable_points($y_p,points,c$)

50:     $ligandablePoints\leftarrow emptyList$

51:     for $point\in points$ do

52:         if $y_p\left[point\right]>c$ then                  ▹ Check if prediction for $point$ is higher than threshold c

53:              append $point$ to $ligandablePoints$

54:         end if

55:     end for

56:     return $ligandablePoints$

57: end function