Charge Group Partitioning

As outlined in section 7.6.3.1, the charge group determination algorithm implemented here has a wparameter to provide a soft upper limit for the size of charge groups. The ATB utilises a more in depth and robust algorithm for determining charge groups which has undergone heavy testing. As a means to determine an optimal value ofwto utilise within CherryPicker, charge groups obtained using various values of ware compared to charge groups obtained by the ATB. The molecules in SRC9064 had charge groups calculated withwvalues ranging from 1 to 10. Three metrics were measured for each charge group obtained: the number of atoms in the charge group, the charge of the charge group, and the diameter of the sphere which would just encompass all the atoms of the charge group. These metrics were also calculated for the charge groups assigned by the ATB. An optimalwvalue would then provide the best overall reproduction of the metrics. Figure 8.2 shows the distributions of charge group sizes, i.e. the number of atoms in the charge group, for eachwvalue. Whenwis less than 4, charge group sizes tend towards much smaller

8. Testing and Discussion

I II III IV V VI VII VIII IX X

Figure 8.1: Structural diagrams of the 23 molecules within CPT023 used to test the CherryPicker pa- rameterisation algorithm. The molecules are: (I) 1-bromohexadecane, SDBS No.: 1743, ATB MolID: 193517; (II) 1-chlorohexadecane, SDBS No.: 5998, ATB MolID: 193519; (III) 1-hexadecanethiol, SDBS No.: 7774, ATB MolID: 193577; (IV) didecyl disulfide, SDBS No.: 7589, ATB MolID: 193574; (V) 2-chloroethyl tetradecanoate, SDBS No.: 5009, ATB MolID: 193544; (VI) 2,4,5-trichlorophenyl

p-(octyloxy)benzoate, SDBS No.: 19281, ATB MolID: 193539; (VII) stearamide, SDBS No.: 7918, ATB MolID: 193566; (VIII) palmitamide, SDBS No.: 7746, ATB MolID: 193569; (IX) hexadecy- lamine, SDBS No.: 1602, ATB MolID: 193552; (X) 3-octadecylthiophene, SDBS No.: 19372, ATB MolID: 193576; (XI) tetrabutylurea, SDBS No.: 22628, ATB MolID: 193564; (XII) 4-bromo-4- heptylbipheny, SDBS No.: 18679, ATB MolID: 194901; (XIII) tris(2-ethylhexyl)amine, SDBS No.: 18489, ATB MolID: 193545; (XIV) ditetradecyl sulfide, SDBS No.: 7756, ATB MolID: 193634; (XV) 2,2(o-phenylenebis(methylenethio))dinaphthalene, SDBS No.: 16336, ATB MolID: 224564; (XVI) 7-tetradecanol, SDBS No.: 41018, ATB MolID: 193571; (XVII)p-(dodecyloxy)nitrobenzene, SDBS No.: 9469, ATB MolID: 193558; (XVIII) retinol, SDBS No.: 22561, ATB MolID: 193555; (XIX) N-lauroyl-N-methylglycine, SDBS No.: 15483, ATB MolID: 193567; (XX) 1,3-bis(1-(2- hydroxyethyl)-4-piperidyl)propane, SDBS No.: 7827, ATB MolID: 193560; (XXI) N,N,N,N- tetrabutyldiethylenetriamine, SDBS No.: 41223, ATB MolID: 193547; (XXII) N-dodecylaniline, SDBS No.: 7904, ATB MolID: 193548; (XXIII) N-(4-hydroxy-3-methoxybenzyl)nonanamide, SDBS No.: 53218, ATB MolID: 193562.

8.1. Algorithm Optimisation

XI XII XIII XIV XV XVI

XVII XVIII XIX XX XXI XXII XXIII

8. Testing and Discussion

Figure 8.2:Distributions of the charge group size, the number of atoms in a charge group, as determined withwvalues from (a) 1 to (j) 10. The filled blue histograms show the size distributions produced for eachwvalue and the green outlines show the reference ATB distribution. The axes are scaled to fit the reference values vertically and the largest calculated value horizontally.

sizes than the reference sizes. Values of between 4 and 6 do a reasonable job of reproducing the reference distribution as they show similar normalised counts to the reference distribution across a broad range of group sizes, whereas values larger than 6 result in a much broader size

8.1. Algorithm Optimisation

Figure 8.3:Distributions of the charge of charge groups as determined withwvalues from (a) 1 to (j) 10. The filled blue histograms show the charge group charge distributions produced by eachwvalue, and the green outlines show the reference ATB distribution. The axes are scaled to show peaks in the distribution at−2,−1 and 1 which would otherwise be overwhelmed by the large peak at 0.

distribution. These results are not entirely unexpected. A large wlimit would tend to favour large charge groups as they can more easily obtain integer charges.

8. Testing and Discussion

figure 8.3. Lowwvalues show a broad peak around a charge group charge of 0 with peaks at−2, −1 and 1 not all being distinguishable untilw=4. Withw=5 or 6, the calculated distributions match the ATB reference distribution very well, thoughw=6 has slightly narrower peaks and the peak at−2 is shifted slightly towards 0 relative to the corresponding reference peak. Larger values ofwresult in narrower peaks than the ATB reference as charge groups tend more towards the expected integer charges.

Even though sharper peaks would be desired in general, as they indicate charge groups closer to the optimum integer charge, they come at a cost of increased charge group diameter, as shown in figure 8.4. Generally speaking, the ATB charge group diameter distribution has a sharp drop off at the upper size limit, whereas the distributions produced by the method described in algo- rithm 3 tend to gradually reduce. This is due to the ATB algorithm including an explicit upper limit to the diameter of a charge group as part of the optimisation method, whereas the algorithm presented here only limits the size of charge groups. The lower limits are identical due to natural constraints imposed by bonded atoms. With awvalue less than or equal to 3, the calculated distributions have 99.9% of the distribution below the reference upper limit. Withw=4, this drops to 95%,w=5 is 84.6% and it continues to rapidly decrease untilw=10 has only 40.7% of charge group diameters below the reference upper limit.

Taking all three metrics into consideration, a value ofw=5 provides the best reproduction of the ATB reference charge groups. The charge group size distribution has a slightly larger pro- portion of medium sized charge groups, and less of the larger sizes. The charge group charges distribution is the best match to the reference overall, narrowly better than w=6 due to the slightly wider peaks, and the location of the peak at−2, and the diameter distribution matches the reference at low diameters before petering off past the upper limit. As such, all parameteri- sations discussed here were performed with aw=5 value.