Systematic errors in parameters

Parameters with sub-populations of errors, suggesting a split could be useful

Bonds

Summary:

• Errors potentially due to electrostatics and/or other parameters: b4 (a41), b8 (elect), b9 (r4 angle), b42 (elec), b56 (a33/r4), b57 (a33/r4), 70 (elec), 73 (elec), 68 (elec) (all halogens)

• Errors likely due to bonds: b6, b10 (could be angle), b17 (could be due to r5 angle), b25, b27, b56

• Unclear: b9, b11, b28, b88

• More data: b24, (not listed here as a problem but has only a few training molecules) b82, b81, b78, 77, b66

Open

The sub-population around 0.05 is largely 5-membered rings with the pattern [r5]=;@[#6X3;r5:1]-;@[#6X3;r5:2]=;@[r5], probably due to angle issues. The sub-population around -0.1 is largely un-protonated carboxyl groups with the pattern [#6X3:1]-[#6X3:2](-[#8X1])=[#8], possibly due to electrostatic/charging issues. Both are separated in MSM distribution also.

Open

Most of the errors around 0.05 is non-ring. MSM shows difference between =;!@ and in-ring.

Angles

Parameters with systematic errors (median error > 0)

Bonds

Bonds with abs(median(length_MM - length_QM)) > 0.025 A

Angles

Angles with abs(mean(MM_angle - QM_angle)) > 2