Chemical Space

Chemical space is the mathematical and conceptual framework in chemoinformatics, medicinal chemistry, materials science, and astrobiology that denotes the multidimensional ensemble spanned by all energetically stable, valence-valid molecular structures and their associated physicochemical property coordinates. By direct analogy to the astronomical cosmos—in which individual stars and galaxies occupy coordinates in physical spacetime—every possible chemical compound occupies a discrete point in chemical space defined by its covalent graph topology, molecular descriptors (such as molecular weight, octanol–water partition coefficient log P, and topological polar surface area), or 2,048-bit molecular fingerprints.

While the total number of chemical substances ever synthesized, isolated, or registered by humanity across the Chemical Abstracts Service (CAS) Registry and PubChem and ChEMBL stands at roughly 2.8 × 10⁸ (280 million) unique structures—and commercial make-on-demand combinatorial libraries such as Enamine REAL and WuXi GalaXi encompass ~4 × 10¹⁰ (40 billion) compounds—these cataloged repositories represent an infinitesimal fraction of theoretically accessible organic chemistry. In 1996, Regine S. Bohacek, Colin McMartin, and Wayne C. Guida calculated that the subspace of small, drug-like organic molecules obeying Lipinski's Rule of Five (up to 30 non-hydrogen heavy atoms composed of carbon, nitrogen, oxygen, sulfur, and halogens) contains between 10⁶⁰ and 10⁶³ distinct molecules, a combinatorial cardinality exceeding the total number of atoms in the Solar System (~10⁵⁷).

Upper-Bound Estimates vs. Synthetically Realizable Drug-Like Space

Because adding each heavy atom to a molecular graph multiplies the number of branching topologies, ring closures, heteroatom substitutions, and stereocenters according to George Pólya's (1937) Enumeration Theorem, the cardinality of chemical space grows according to an exponential scaling law (N_molecules ∝ e^α*N*_heavy = b^N_heavy, where base b = e^α ≈ 2.2 to 8.0 depending on valence and ring strain constraints):

• The Bohacek–McMartin–Guida Estimate (~10⁶⁰–10⁶³): Derived in 1996 by combinatorially decorating branched carbon skeletons of ≤30 heavy atoms (MW ≤ 500 Da) with N, O, S, and halogen atoms while filtering out hydrolytically unstable functional groups (such as peroxides, hemiaminals, and anhydrides). Synthesizing just one single molecule of each of these 10⁶⁰ compounds would require ~10³⁵ kilograms of carbon—roughly 10⁵ solar masses.
• The Polishchuk–Madzhidov–Varnek Realistic Estimate (~10³³): In 2013, Pavel G. Polishchuk and colleagues restricted combinatorial assembly to experimentally validated medicinal chemistry fragments and reaction connection rules observed in literature syntheses, estimating that ~10³³ drug-like molecules are practically synthesizable under standard laboratory conditions.

Jean-Louis Reymond's GDB-11, GDB-13, and GDB-17 Enumerations

To chart small-molecule chemical space systematically without bias toward historically synthesized scaffolds, Jean-Louis Reymond and coworkers at the University of Bern constructed the Generated Databases (GDB) by starting from mathematical hydrocarbon graphs (GENG / nauty) and exhaustively substituting bonds and heteroatoms while enforcing valence rules, ring strain limits (excluding fused cyclopropanes, cyclobutanes, and bridgehead double bonds violating Bredt's Rule), and functional-group stability in water at pH 7.4:

1. GDB-11 (Fink & Reymond, 2007): Enumerated all 2.64 × 10⁷ (26.4 million) stable organic molecules containing up to 11 heavy atoms of C, N, O, and F (MW ≈ 150–180 Da).
2. GDB-13 (Blum & Reymond, 2009): Enumerated 9.77 × 10⁸ (977 million) stable organic molecules up to 13 heavy atoms of C, N, O, S, and Cl, representing the fragment-like chemical space governed by Congreve's Rule of Three.
3. GDB-17 (Ruddigkeit, van Deursen, Blum, & Reymond, 2012): Enumerated 1.664 × 10¹¹ (166.4 billion) organic molecules containing up to 17 heavy atoms of C, N, O, S, F, Cl, Br, and I (MW ≤ 250 Da). Comparison of GDB-17 against PubChem revealed that >99.8% of mathematically stable 17-atom scaffolds—particularly three-dimensional sp³-rich bridged and spirocyclic rings—have never been synthesized by chemists, who historically favored flat, sp²-aromatic biaryl rings amenable to Suzuki–Miyaura palladium cross-coupling ("escape from flatland", Lovering et al., 2009).

PCA, t-SNE, UMAP, and Tree-Maps (TMAP)

Because a molecule represented by a 2,048-bit ECFP4 molecular fingerprint or 42 continuous physicochemical descriptors resides in a 42- to 2,048-dimensional vector space, visualizing chemical library diversity and QSAR applicability domains requires dimensionality reduction into 2D or 3D maps:

• Principal Component Analysis (PCA) and ChemGPS: Projects high-dimensional standardized descriptor matrices onto orthogonal linear eigenvectors maximizing variance. Tudor Oprea and Johan Gottfries (2001) formulated ChemGPS (Chemical Global Positioning System), anchoring drug-like space along principal axes of molecular size, lipophilicity (cLog*P*), polarizability, and flexibility, while Reymond's MQN (Molecular Quantum Numbers) projects databases using 42 fast atom/bond/ring counts.
• Non-Linear Manifold Projections (t-SNE, UMAP, and GTM): t-Distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP) preserve local Tanimoto neighborhood topology, clustering structural analog series into discrete islands. Generative Topographic Mapping (GTM) fits a probabilistic bidirectional manifold capable of projecting new external compounds without recomputing the entire embedding.
• TMAP (Tree-Map, Probst & Reymond, 2020): Uses MinHash Locality-Sensitive Hashing (LSH) forests and Kruskal's minimum-spanning-tree algorithm to lay out millions of molecules as an interactive 2D tree graph where every branch represents a continuous structural transition.

Beyond pharmaceutical discovery, chemical space provides a quantitative metric for distinguishing abiotic prebiotic chemistry from living metabolomics and biosignatures. Carbonaceous chondrite meteorites (such as the Murchison meteorite) contain tens of thousands of low-molecular-weight abiotic organic isomers generated by random radical reactions, whereas terrestrial biochemistry utilizes a sparse, highly conserved island of ~10³ primary metabolites (such as ATP, the 20 proteinogenic amino acids, and citric acid cycle intermediates) alongside ~2 × 10⁵ secondary natural products.

In 2021–2023, Leroy ("Lee") Cronin and Sara Imari Walker introduced Assembly Theory, which defines the Molecular Assembly Index (MA) of a molecule as the minimum number of recursive bond-joining steps required to construct its molecular graph from basic atomic building blocks while reusing previously constructed subgraphs. Because combinatorial chemical space explodes above N_heavy ≈ 15, random abiotic chemistry without evolutionary memory or enzymatic templates cannot produce detectable concentrations of molecules with MA ≥ 15 (measurable directly by tandem fragmentation mass spectrometry), establishing high-assembly regions of chemical space as an agnostic planetary biosignature.

How did Bohacek, McMartin, and Guida arrive at the famous 10⁶⁰ estimate for drug-like chemical space?

In their 1996 Medicinal Research Reviews analysis, Regine Bohacek, Colin McMartin, and Wayne Guida asked how many distinct, chemically stable organic molecules below 500 Daltons could fit inside a protein active site. Starting from branched and cyclic hydrocarbon skeletons of up to 30 heavy atoms, they calculated the number of ways to introduce unsaturations (double/triple bonds), stereocenters, and heteroatom substitutions (N, O, S, F, Cl, Br) at every position while pruning unstable linkages (such as O–O peroxides or geminal diols). Multiplying the independent combinatorial choices across 30 heavy-atom vertices yielded an upper bound of 10⁶⁰ to 10⁶³ structures.

What is "Escape from Flatland" in the context of chemical space and GDB-17?

In a 2009 paper titled "Escape from Flatland", Frank Lovering, Jack Bikker, and Christine Humblet demonstrated that clinical drug candidates have a higher Fraction of sp³-hybridized carbons (Fsp³ = N_{sp3 carbons} / N_{total carbons}) and more chiral centers than early-stage discovery hits, which are skewed toward flat sp² aromatic rings (Fsp³ < 0.25) because of synthetic reliance on palladium-catalyzed cross-coupling reactions. When Jean-Louis Reymond's GDB-17 database enumerated 166.4 billion stable 17-atom molecules, it proved that unbiased mathematical chemical space is dominated by three-dimensional, non-aromatic spirocycles and bridged rings (Fsp³ > 0.60) that improve aqueous solubility by disrupting flat π-stacking crystal lattices.

How does Assembly Theory use chemical space to detect extraterrestrial life?

Most astrobiological biosignatures look for specific terrestrial biomolecules (such as homochiral L-amino acids or membrane hopanoids), which could miss alien life built from alternative biochemistry. Assembly Theory (Cronin & Walker, 2021) instead measures where a sample sits in combinatorial chemical space via its Molecular Assembly Index (MA)—the shortest pathway of subgraph duplications needed to build the molecule, which correlates linearly with the number of fragment peaks observed in tandem mass spectrometry (MS/MS). Because the number of possible isomers exceeds 10²⁰ at high complexity, finding any molecule in macroscopic abundance (>10⁶ copies) with MA ≥ 15 requires directed selection and information-driven catalysis.

1. Bohacek, R. S., McMartin, C., & Guida, W. C. (1996). The art and practice of structure-based drug design: A molecular modeling perspective. Medicinal Research Reviews, 16(1), 3–50. https://doi.org/10.1002/(SICI)1098-1128(199601)16:1<3::AID-MED1>3.0.CO;2-61098-1128(199601)16:1%3C3::AID-MED1%3E3.0.CO;2-6)
2. Ruddigkeit, L., van Deursen, R., Blum, L. C., & Reymond, J. L. (2012). Enumeration of 166 billion organic small molecules in the chemical universe database GDB-17. Journal of Chemical Information and Modeling, 52(11), 2864–2875. https://doi.org/10.1021/ci300415d
3. Lovering, F., Bikker, J., & Humblet, C. (2009). Escape from flatland: Increasing saturation as an approach to improving clinical success. Journal of Medicinal Chemistry, 52(21), 6752–6756. https://doi.org/10.1021/jm901241e
4. Marshall, S. M., Mathis, C., Carrick, E., Keenan, G., Cooper, G. J. T., Graham, H., ... & Cronin, L. (2021). Identifying molecules as biosignatures with assembly theory and mass spectrometry. Nature Communications, 12(1), 3033. https://doi.org/10.1038/s41467-021-23258-x