Retrosynthetic Analysis

Why retrosynthesis matters

• Transformed synthesis into a discipline. Before Corey, synthesis was apprenticed intuition — a student watched a master and absorbed taste. After 1969, retrosynthetic analysis turned route design into an explicit, teachable, debuggable problem-solving framework with rules, conventions, and notation. Today every undergraduate organic class teaches it.
• Convergence is the yield multiplier. A 10-step linear synthesis at 80% per step gives ~11% overall; a 10-step convergent (5+5) gives ~21%, twice as much. Lipitor (atorvastatin, Pfizer, $13B/yr peak) is made convergently — pharmaceutical process scientists pick convergent routes specifically to halve cost of goods.
• LHASA was the first synthesis-AI. Corey's 1969 LHASA program pre-dated computer chemistry by decades. It encoded ~250 transforms and could propose multi-step disconnections for steroid-class targets. Today's AI descendants (Synthia, IBM RXN, AiZynthFinder) train on 5-50 million literature reactions and propose viable routes for ~70% of benchmark drug targets.
• Scaffold-based drug discovery. Pharmaceutical companies maintain in-house libraries of 10-100 million compounds; retrosynthesis is used in reverse to suggest which scaffolds can be made from which available building blocks, mapping the "synthesizable space" of a project. Pfizer, Merck, and Novartis each run dedicated retrosynthesis teams of 5-20 chemists.
• Total synthesis benchmarks. Robert Burns Woodward's strychnine (1954, 28 steps), R.B. Woodward's vitamin B12 (1973, 100 steps with Eschenmoser), and K. C. Nicolaou's Taxol (1994, 51 steps) are landmark retrosyntheses that showed the framework scales to molecular complexity. Strychnine has been re-synthesized 16+ times since, each in fewer steps as retrosynthetic understanding improved.
• Process chemistry cost reductions. A successful retrosynthetic redesign can cut step count from 12 to 6, dropping cost of goods by 5-10x. The Sitagliptin (Januvia, Merck) process chemistry route was redesigned via biocatalytic retrosynthesis to remove a Rh-catalyzed step, saving Merck ~$50M/yr in 2010.
• Teaches polarity matching. Disconnection forces the chemist to think in synthon terms — every bond-forming reaction in the forward direction has a polar logic (electrophile + nucleophile, or radical pair). Retrosynthesis exposes this logic and trains chemists to predict reactivity even for unfamiliar substrates.

Common misconceptions

• Disconnect any bond and call it retrosynthesis. No — disconnections must yield synthons whose synthetic equivalents are realistic. Disconnecting a C-C bond between two saturated carbons of equal substitution is rarely useful because no realistic reagent corresponds to a saturated carbanion or carbocation. Strategic bonds are typically those flanked by heteroatoms (C-C alpha to C=O) or in rings.
• The shortest synthesis is always best. Not at scale. A 12-step convergent route with cheap, non-toxic reagents and no chromatography can beat a 6-step linear synthesis that requires Pd at 5 mol% and a chiral auxiliary. Process chemistry at scale optimizes cost per kg, not step count.
• Synthons must always be ions. Radical synthons are equally valid. The methyl radical CH3• is a synthon for a methyl reagent in radical chemistry; AIBN-initiated radical addition is the synthetic equivalent. Modern photoredox catalysis has expanded the radical-synthon palette dramatically (since ~2010), and retrosynthesis tools like IBM RXN now include radical disconnections.
• Functional groups don't count as disconnections. They do — functional-group interconversions (FGI) are first-class retrosynthetic moves, drawn with a different arrow. An alcohol can be retrosynthetically traced to an aldehyde via NaBH4-reduction FGI, even though no C-C bond is broken. FGI mastery is what unlocks disconnections that would otherwise require impossible reagents.
• You disconnect in the order you'll synthesize. Reverse — retrosynthesis goes target → starting materials. Synthesis goes starting materials → target. This is the entire point of the framework: planning happens in reverse, execution happens in forward.
• AI retrosynthesis is a solved problem. Even the best 2024 tools (Synthia, IBM RXN) match expert chemists on ~40-70% of test cases but underperform on stereochemistry, ring-strain handling, and convergence-aware planning. Human expert taste is still required for non-trivial targets.

The disconnection workflow

The retrosynthetic workflow has four stages. Stage 1: Analyze the target. Identify functional groups, stereogenic centers, ring systems, and connectivity. The chemist annotates the target with key structural features and notes any symmetry that suggests a midpoint disconnection. Stage 2: Identify strategic bonds. These are bonds whose disconnection maximally simplifies the target — typically bonds in rings (especially newly closed rings), bonds adjacent to functional groups that can act as handles, and bonds at the molecular midpoint that yield two roughly equal precursors. Corey enumerated five quantitative rules for strategic-bond selection.

Stage 3: Disconnect each strategic bond and write the synthons. The retrosynthetic arrow ⇒ is drawn pointing from the target to the precursors. Each disconnection produces two (or more) synthons with assigned polarities — typically one cation (synthetic equivalent: electrophile) and one anion (synthetic equivalent: nucleophile). The chemist writes the real synthetic equivalent below each synthon: e.g., the acetyl cation CH3CO⁺ corresponds to acetyl chloride or acetic anhydride; the methyl carbanion CH3⁻ corresponds to MeMgBr or MeLi. Stage 4: Recurse. Treat each precursor as a new target, identify its strategic bonds, disconnect, and continue until all branches reach commercial starting materials. Multiple routes are typically generated and compared on overall yield, step count, convergence, cost, and safety.

Functional-group interconversion (FGI) supplements the disconnection toolkit. When the target's functional group makes the desired disconnection impossible (e.g., a carboxylic acid where you wanted to disconnect at an aldehyde-equivalent step), the chemist applies an FGI arrow to swap the functional group for a more favorable one. Each FGI corresponds to a known forward reaction (NaBH4: aldehyde → alcohol; DIBAL-H: ester → aldehyde; PCC: alcohol → aldehyde). Iterating disconnections and FGIs builds the complete retrosynthetic tree, which is then translated to the forward synthesis plan with each retro-arrow becoming a forward step in reverse order.

Convergent vs linear synthesis — yield and cost comparison

Synthesis style	Step count	Per-step yield 80%	Per-step yield 90%	Cost-of-goods scaling	Best use case
Linear (single chain)	5 steps	0.8⁵ = 33% overall	0.9⁵ = 59% overall	Linear in steps × material price	Simple targets, <5 steps
Linear	10 steps	0.8¹⁰ = 11%	0.9¹⁰ = 35%	Material accumulates through chain	Rare for >6 steps in pharma
Linear	20 steps	0.8²⁰ = 1.2%	0.9²⁰ = 12%	Prohibitive at scale	Academic curiosities only
Convergent (2 branches)	10 steps (5 + 5)	0.8⁵ × 0.8⁵ × 0.8 = 21%	0.9⁵ × 0.9⁵ × 0.9 = 31%	~½ of linear cost	Pharma APIs (Lipitor, Crizotinib)
Convergent (3 branches)	15 steps (5 + 5 + 5)	~17%	~26%	~⅓ of linear cost	Complex natural products
Iterative coupling	10 building-block additions	0.95¹⁰ = 60% (peptide-grade)	0.99¹⁰ = 90% (DNA synthesis)	Linear + automation	Peptides, oligonucleotides, MIDA boronates
Total synthesis (Woodward strychnine 1954)	28 linear	~0.0001% reported	—	Heroic; not for production	Methodology development
Modern Vanderwal strychnine 2011	6 steps	~13% overall	—	1000× simpler than 1954	Demonstrates retrosynthetic progress

Famous syntheses

• R. B. Woodward's strychnine (1954, 28 steps). The benchmark that demonstrated total synthesis of a complex alkaloid was possible. Woodward chose his disconnections by intuition; modern retrosynthesis (Vanderwal 2011, 6 steps) cuts the route by 4×. Corey explicitly cited Woodward's syntheses as the inspiration for formalizing retrosynthetic logic.
• R. B. Woodward and Albert Eschenmoser's vitamin B12 (1973, ~100 steps). The most ambitious synthesis ever attempted, requiring 100 graduate students over 11 years. Demonstrated that retrosynthetic disconnection could plan a target with 9 stereocenters and a unique cobalt-corrin macrocycle. Many of the disconnections (the Eschenmoser sulfide contraction, A-B ring annulation) became standard methodology.
• K. C. Nicolaou's Taxol (1994, 51 steps). Anticancer drug originally isolated from Pacific yew bark. The retrosynthetic plan disconnects the strained 8-membered B ring as the strategic bond, with two convergent fragments meeting at a key Heck or McMurry coupling. Total synthesis demonstrated that natural products could be re-engineered when supply was limited.
• E. J. Corey's prostaglandin syntheses (1969-1980s). Corey's group built nearly all major prostaglandins from a common bicyclic intermediate (the "Corey lactone") via convergent disconnections. This showcased the power of identifying a single strategic bicyclic precursor and elaborating outward — a unifying retrosynthetic insight that produced ~50 different prostaglandins from one platform.
• Industrial Lipitor (atorvastatin, Pfizer). The world's best-selling drug ($13B/yr peak). Process chemistry uses convergent retrosynthesis: the chiral side-chain (4 stereocenters) is built enzymatically in one branch; the heterocyclic core in another; they unite via Paal-Knorr and Knoevenagel-Doebner condensations. Cost of goods at multi-tonne scale <$100/kg.