Network anchoring

Warning: This example is for ARIA >= 2.2 only.

Background

The network-anchoring approach was proposed by ((Herrmann et al. 2002) in order to reduce the number of possibilities of NOE peak assignment during the iterative NMR structure refinement with CANDID and CYANA (G ̈untert, 2004). The approach is based on the ranking of each NOE assignment, using the information about the assignments of neighbourig nuclei in the 3D space. It is based on the hypothesis that the correctly assigned restraints form a self- consistent subset of the network of restraints. The network-anchoring approach is new in ARIA, as, in previous versions, it was not present, nor in the python code, neither in the CNS Fortran code. It was implemented in the following way, which is an adaptation of (Herrmann et al., 2002) to the ARIA philosophy. The network anchoring is indeed used to calculate a contribution score, which replaces the usual contribution weight of the ARIA protocol.

A preliminary analysis over all amino-acids and possible sequences of two amino-acids determined the list of spins pairs separated by at most two dihedral angles, and whose the distance is smaller than 5.5 A, whatever are the dihedral angle values. The analysis was not performed using a geometric formal calculation as previously done in the literature (W ̈uthrich et al., 1983), but by an exhaustive sampling of possible conformations of the amino-acids and sequences of two amino-acids, using the CNS topology definition and only allowing dihedral angle rotations. All the possible ”covalent” distances are stored in a XML file.

At the beginning of the ARIA calculation, the network is created: for each pair of spins α, β, it contains all protons γ of the residue range [r − 1,r + 1],
conneted to either α or β with an initial assignments, and where r is the residue number of the spin α or β. For a given pair of spins a, b belonging to the α, β network, the covalent score describes the covalent structure between these spins. If a,b belongs to the list of spins pairs previously determined by the exhaustive sampling of possible amino-acid conformations, the covalent score V_ab^(cc) is:

Vcc

The parameters V_max and V_minvalues are built-in and set to values 1.0 and 0.1, respectively.

The number of indirect connections between the spins α and β via the atoms γ belonging to the α, β network is calculated as:

(4)

where v_αγ = max(V _αγ^(cc) , V_αγ^vol) if max(V_αγ^(cc) , V _αγ^vol) > V_min or 0.0 otherwise.

V _αγ^volis the volume of the corresponding NOE assignments between the spins α and γ.

The network score N_c for each contribution c is calculated as the mean of the values N_αβ obtained (Eq. 4) for each spin pair α, β included into the contribution. Finally, the contribution scores Nc are normalized.

From the network contribution score N_c , the contribution score Sc , which will be used then as the contribution weight, is obtained from the product of the value N_normc and the weight w_cdetermined the distance observed in the structures calculated in the previous iteration. This product is also normalized:

The contribution weigths are updated with the new score values S_c . Scores S_res and S_atodepending on the residues and atoms are then derived from the contributions scores N_c , in order to apply criteria allowing to keep or to reject an AriaPeak. First, the scores N_care summed over each residue pairs involved in this a peak to obtain residue score S_r. For each AriaPeak, a residue score S_res and an atom score S_atoare calculated:

SresSato

where r are the residue pairs and c the contributions involved in the peak, and n_sp is the number of spin pairs.

Using the parameters S_resand S_atom , the peak is declared active and therefore conserved if one of the two following rules is verified:

S_res ≥ N_high ^res

S_res ≥ N_min ^res && S_atom ≥ N_min ^ato

Thus, the network-anchoring is used to

Remove possible noise peaks which have low S_res and S_atomscores.

Reduce the ambiguity within each peak by eliminating contributions with low weight S_c

The two rules previously described are corresponding to the rules proposed in the original publication about network-achoring (Herrmann et al, 2002). In CANDID (Herrman et al, JMB 2002), a peak p is kept if one of the two following conditions (page 215, ``Elimination of spurious NOE cross-peaks'') is verified:

< \overline{N} >_p >= \overline{N}_high
< \overline{N} >_p >= \overline{N_min} && <N>_p >= N_min

where \overline{N} stands for a overlined N.

The value of \overline{N}_high is constant and equal to 4.0 (Table 4, page 225, Herrman et al, JMB 2002) and the values of \overline{N_min}
and N_min vary (Table 5, page 225, Herrman et al, JMB 2002). For a run with 7 iterations, \overline{N_min} is equal to 1.0 at the first iteration, equal to 0.75 at the second iteration and equal to 0.5 in the following iterations. In the same run, N_min is equal to 0.25 in the two first iterations, and is then set to 0.4.

The parameters N_high ^res, N_min ^res and N_min ^ato of ARIA correspond thus to the parameters \overline{N_high}, \overline{N_min} and N_min in CANDID.

Example description

One example of the use of the network-anchoring procedure is given for the Tudor domain (ref):

project xml file: werner_network.xml
project directory: run_net
run archive: network_anchoring.tgz