A new approach for coverage of the conformational space by a limited number
of conformers is proposed. Instead of using a systematic search whose time
complexity increases exponentially with degrees of freedom, a genetic algo
rithm (GA) is employed to minimize 3D similarity among the conformers gener
ated. This makes the problem computationally feasible even for large, flexi
ble molecules. The 3D similarity of a pair of conformers is assumed to be r
eciprocal to the root-mean-square (rms) distance between identical atomic s
ites in an alignment providing its minimum. Thus, in contrast to traditiona
l GA, the fitness of a conformer is not quantified individually but only in
conjunction with the population it belongs to. The approach handles the fo
llowing stereochemical and conformational degrees of freedom: rotation arou
nd acyclic single and double bonds, inversion of stereocenters, flip of fre
e corners in saturated rings, and reflection of pyramids on the junction of
two or three saturated rings. The latter two were particularly introduced
to encompass the structural diversity of polycyclic structures. However, th
ey generally affect valence angles and can be restricted up to a certain le
vel of severity of such changes. Stereochemical modifications are totally/s
electively disabled when the stereochemistry is exactly/partially specified
on input. Three quality criteria, namely robustness, reproducibility, and
coverage of the conformational space, are used to assess the performance of
various GA experimental settings employed on four molecules with different
numbers of conformational degrees of freedom. It was found that with the i
ncrease of the ratio between the number of parents and children, the reprod
ucibility of GA runs increases whereas their robustness and coverage decrea
se. Force field optimization of conformers for each generation was found to
improve significantly the reproducibility of results, at the cost of worse
conformational coverage.