Actually most of the times you dont know exactly about the possible properties of that substance, but when you know several hundred different molecules in structure and effects, and what else is chemical related to it, will give you the ability to predict its assumed effects more or less exactly, especially when someone is familiar with that sort substances. It is however mostly estimated.
Also a lot of people may have considered to synthesize some as RC known substances or have even attempted it. This involves reading many papers from the scientists that discovered it first, and those guys usually will test their molecules for the affinity (how strong it will bind) on the receptor types their are desired to interact with. This is where also many valuable facts that express the potency with a value, which makes it easier to compare too. But most important, you can see how a small change in structure will alter the substances effects. So one should be able to estimate the effects to some extent after he had studied enough properties to compare a lot of related substance.
The term Structure-activity-relationship (SAR) is used then, when so much molecules of a type are made, that one can to some degree predict how strong a still unsynthesized compound will interact with its specific receptors.
At least thats how methoxetamine was made, considering that a N-ethyl group is the most potent substitution at the aminogroup, and the fact that a MeO-group on position 3 on the benzene ring would increase its potency several times over that of the substance it arises from, PCE.
I think i read somewhere that structure activity relationship has in fact to get calculated with a computer, as it is so complex and one has to deal with several unknown values. That is also the reason that first a lot of compounds, as much as possible, have to be made so it is very likely that you will discover some of these constants to have at least a vague orientation what parts of that molecule determine its effects.
The SAR will be more exactly if one eliminates as much unknown values as possible. But... im not really sure if thats how it actually works.

Edit: Oh I forgot to add, a drug has to share essential structural similarities to the corresponding neurotransmitter that it acts on mainly. For example, nearly every upper has the backbone of adrenaline and dopamine (both very similiar phenethylamines) in it, most hallucinogens have a tryptamine containing core, like our own serotonine (which is of course the receptor where their action will take its major part).
But in this sector there is no exactly predictable SAR, there are frequently unexpected relations discovered. Our endocannabinoids are a good example too, they are just long fatty acids without any ring in it.
Our cannabinoid receptors although have a shape that looks like it is designed for molecules with thc-similiar shape. But those long chains of endocannabinoid will fold itself in a thc-like shape when they dock at the receptor site, totally unexpected. Thats also the reason i believe that took them so long to identify them as actual essential neurotransmitter.

Edit2: cleaned that crappy grammar out, i should only post when sober^^