Retrosynthesis is the subtle art of constructing simple starting materials from a target molecule using known or mechanistically viable reverse reactions, or "transforms." By sequentially applying transforms to a target molecule, the target is reduced step-wise into simpler and simpler starting materials until at last a commercially available compound is reached.
The seminal text on retrosynthesis is Corey and Cheng's The Logic of Chemical Synthesis, which describes five several retrosynthetic strategies and provides a number of examples to illustrate these strategies.
- Transform-based Strategies: These rely on the application of a key structurally simplifying transform that reduces a molecule's complexity immensely in a single step. Steps on either side of the key transform work to build up the "retron," or keying molecular structure, for the central transform.
- Structure-based Strategies: These focus on simplifying the target compound to a particular structure. A number of natural products are derived from common, widely available commercial materials like amino acids, carbohydrates, small-molecule hormones, and the like.
- Functional-Group-based Strategies: These center on the addition, interconversion, and deletion of particular functional groups as a way to modulate complexity in the target. Oftentimes a target differs from a key retron by only a few functional groups; transforms that can add these functional groups without disturbing the rest of the molecule are the foundation of functional-group-based strategies.
- Topological Strategies: These center on transforms that simplify the target topologically by breaking, for example, a key bond contained in multiple rings or a bond between a core ring system and an appendage. Thus in the forward direction, good topological transforms increase topological complexity substantially in one step.
- Stereochemical Strategies: These make use of stereospecific or stereoselective transforms in order to establish the target in enantiopure form.
In practice, all five of the strategies have to be applied at once in order to reduce a target structure in the most efficient way possible. This makes retrosynthesis, frankly, a bitch at times. Not knowing enough reactions to perform key bond disconnections (bond-forming reactions in the forward direction) is one thing, but knowing and just not seeing them is another thing entirely. The latter is the pitfall that affects most early chemistry graduate students. And even if you can see retrons in a target structure easily, care must be taken to not back yourself into a corner where selectivity in the forward direction is impossible.
A good example where this comes into play is the Wittig reaction, which in the reverse direction transforms a double bond into a ketone/aldehyde and an alkyl halide. If the target has a double bond and a ketone in it, disconnecting the double bond to a ketone via the Wittig transform is a bad idea, because in the forward direction the other ketone would react just as well as the one you want to turn into a double bond. Unless there is some glaring steric issue, selectivity has become impossible. This, to me, illustrates one of the general rules of retrosynthesis: never disconnect into a reactive functional group already present in the target.
If you have ambitions of becoming a chemistry graduate student, retrosynthesis is one of the hardest things you'll ever do. That said, it's far from impossible! Being very comfortable with reactions in both the forward and reverse directions is essential, and looking at literature retrosyntheses on a regular basis can also help. Be wary of current papers though, as these often use new, untested reactions very specific to the substrates they're dealing with. With practice, anyone can be a retrosynthetic pro!