Algorithm 1: Acmi-PF's algorithm for growing a protein model.

input : density map y, amino-acid marginals p̂k(bk)

output: set of protein models $x_{1:K}^{(i)}$ and weights $w_K^{(i)}$

// start at some AA with high certainty about its location

choose k such that p̂k( $b_k^{(i)}$) has minimum entropy

foreach particle i = 1… N do

choose $b_k^{(i)}$ at random from p̂k( $b_k^{(i)}$)

$w_k^{(i)}$ ← 1/N

end

foreach residue k do

foreach particle i = 1… N do

// choose bk+1 (or bk−1) given $b_k^{(i)}$

{ $b_{k+1}^{\ast m}$ } ← choose M samples from φadj( $b_k^{(i)}$, bk+1)

w*m ← belief p̂i ( $b_{k+1}^{\ast m}$)

$b_{k+1}^{\left(i\right)}$ ← choose $b_{k+1}^{\ast m}$ with probability ∝ w*m

$w_{k+1}^{\left(i\right)}$ ← $w_k^{\left(i\right)}$ · ${\sum }_{m=1}^M$ w*m

// choose skgiven $b_{k-1:k+1}^{(i)}$

{ $s_k^{\ast l}$} ← sidechain conformations for amino-acid k

$p_{\text{null}}^{\ast l}$ ← prob cc( $s_k^{\ast l}$, EDM[bk]) occurred by chance

sk ← choose $s_k^{\ast l}$ with probability ∝ 1/ $p_{\text{null}}^{\ast l}$ − 1

$w_{k+1}^{\left(i\right)}$ ← $w_k^{\left(i\right)}$ · ${\sum }_{l=1}^L$ 1/ $p_{\text{null}}^{\ast l}$ −1

end

end