Chemistry Reference
In-Depth Information
2. Identification of and search for equivalent atoms. An automatic protocol scans
the reference structures and collects equivalent atoms using different structure
descriptors and similarity measures. These descriptors include the atomic num-
ber, the local connectivity graph and bond-order/aromaticity indices. Based on
the statistical agreement of similarity measures, a decision is made for each atom
whether it belongs to an existing group or it is to be considered as a new type.
For each atom, grouped in such a way, the descriptors are stored together with a
code referring to the molecule the atom originates from.
3. Ab initio electronic structure calculations on reference molecules containing
atoms picked up by the above protocol. The molecular densities, used to
construct the current databank, were generated at the B3LYP/aug-cc-pVTZ
level.
4. Calculation of the RDFs. For each type of SPAs, the RDFs are calculated from
the molecular wave functions on a fine radial grid. This step also performs a
statistical analysis and the fit to obtain the analytic SPA-RDFs of the average and
correction densities.
5. Building the ED of the target molecule of a known structure. Input atomic
positions and specific criteria for matching similarity indices are used by the
builder to construct both the average density (
r
X
) and the RDFs of the first-order
correction. The total charge is carried by the average term whose monopole
populations are scaled so as to re-establish the electro-neutrality.
The applicability of the SPA databank is being explored using both simu-
lated and experimental structure factors for molecules of 60-70 atoms, which
can still be challenging targets of X-ray charge density analyses. According to
(19), the library-based SPA protocol keeps the transferable density fixed and
completes the data analysis in terms of the nontransferable part. Since this
option is not yet implemented into XD, our preliminary studies used only the
averaged SPA component. Results of the interpretation of theoretical (B3LYP/
cc-pVTZ) and experimental data (100K, synchrotron radiation, at 1.25
˚
1
resolution [
70
]) of the
Terbogrel
(C
23
H
27
N
5
O
2
) molecule (Fig.
7d
)arepre-
sented here as an example.
The structure factors calculated with the average SPAs drawn from the databank
are in an excellent agreement (
R
0.52%) with those generated from the wave
function. Figure
7a
displays the direct-space residual density for the guanidine
fragment, which exactly maps features unaccounted for by the zero-order (average)
database model. This is the section of the residual ED where the worst agreement
between the “exact” and the library densities was found, as seen by the pronounced
residual contours around the N and C-atoms of the -N-C
¼
N group. The refinement
of multipole populations of the average density lowers the
R
-value to 0.38% and
removes the highest residual contour lines (Fig.
7b
). No bonding features higher
than 0.05 e
˚
3
are left. While the accuracy of the fitted average density is quite
satisfactory, the remaining density features can be almost completely removed by
adjusting the multipole populations of
d
X
½
E
to the residual structure factors
ð
F
ðr
C
Þ
F
ðrÞÞ
. The ADP refinement of the average SPAs against the experimental
