# Simple QM9 properties for prediction

I was looking into an article (doi: 10.1063/1.5058063) which presented a histogram of radius of gyration for samples of different configurations of alanine dipeptide. This property is fairly simple to calculate using $$a_{\mathrm{Rg}}(x)=\sqrt{\frac{\sum_{p} m_{p}|| x_{p}-x_{\mathrm{COM}}||^{2}}{\sum_{p} m_{p}}}$$ The sum considers all atoms $$p = 1, \dots, P$$ of the peptide, where $$m_{p}$$ and $$x_{p}$$ denote the mass and the coordinates of each atom, respectively. $$x_{\mathrm{COM}}$$ denotes the center of mass of the peptide.

In a different problem, I have a number of molecular graphs generated by a network trained using QM9 dataset and am looking for properties that are readily calculable for the newly generated molecular graphs using such packages as RDKit and OpenBable.

The properties in the QM9 dataset itself, like LUMO energy and energy of atomization, need DFT calculation, which is not something that I look forward to do. An example of a property of interest is QED, which can be easily calculated using RDKit package. But QED is more of interest in ZINC dataset.

I'd appreciate comments on any easy to compute property of interest to the community for QM9 dataset.

• Suggestion: Add a bit about your infrastructure, what program do you typically use, do you work on your own or on a larger computer cluster? Equally as a suggestion: please add a link to the publication you refer to, e.g. the doi, because there are already some posts here including histograms about properties in data sets. – Buttonwood Jan 22 '20 at 20:06
• Thanks for the suggestion @Buttonwood. I added doi for the paper, but I'm afraid that the program that I use and my infrastructure would be irrelevant to the question. Not that it is a well kept secret, but it won't help with the problem. – Blade Jan 22 '20 at 20:11

I found 3 properties:

1. The molecular weight (MolWt)
2. Wildman-Crippen partition coefficient (LogP)
3. Topological Polar Surface Area (TPSA)