14 · Carbon and life
Of the hundred-odd kinds of atom, one builds almost all the molecules of life: carbon. And life itself is chemistry that refuses to reach equilibrium. This last chapter puts both ideas together, using nearly everything from the chapters before.
Why carbon?
Carbon has four outer electrons and makes four bonds, pointing to the corners of a tetrahedron (chapter 6). Its bonds to itself are strong, about 350 kJ/mol for C–C, and about as strong as its bonds to hydrogen, oxygen and nitrogen. So carbon can form long chains, branches and rings without any one arrangement being overwhelmingly preferred. Those bonds are also strong enough to survive room temperature for millions of years (chapter 10's barriers).
Silicon, just below carbon in the table, also makes four bonds, but its bonds to itself are much weaker than its bonds to oxygen. In an oxygen-rich world silicon ends up locked in rock (silicates, sand, quartz) rather than in long chains.
The result is an explosion of possibilities. Chemists have catalogued well over a hundred million carbon compounds. Even the simplest family, chains of carbon with hydrogen filling every spare bond (the alkanes), branches combinatorially:
This counts only "constitutional" isomers, meaning different bonding graphs. From 7 carbons on, some skeletons also come in mirror-image pairs (left- and right-handed forms), which the graph picture can't distinguish (chapter 9's caveat). Real chemistry multiplies the count further with double bonds, rings and other elements.
Each of those skeletons is a different substance with its own boiling point. Add functional groups (an –OH makes an alcohol, a –COOH an acid like chapter 12's vinegar, an –NH2 a base) and each skeleton becomes a scaffold for many more. Functional groups are the reusable modules of organic chemistry: an –OH group behaves much the same whatever it's attached to.
The molecules of life
Every big molecule in your body is built from a small kit of parts, using ideas from earlier chapters:
- Proteins are chains of 20 kinds of amino acid. A chain folds into a precise 3-D shape, mostly driven by the hydrophobic effect (chapter 8): oily side groups hide inside, water-loving ones face out. The shape is the function: enzymes (chapter 10's catalysts) are proteins with pockets shaped to hold one reaction's transition state.
- DNA is two chains held together by hydrogen bonds between paired bases (chapter 6): A pairs with T using two hydrogen bonds, and G pairs with C using three. Those bonds are weak enough to unzip for copying, and the strong covalent backbone keeps the message intact.
- Membranes are double layers of soap-like molecules (chapter 8) with oily tails inside and water-loving heads outside. They self-assemble, because that arrangement is the free-energy minimum.
- Sugars are fuel and building material: carbon chains studded with –OH groups, which is why they dissolve so well in water.
Life is not at equilibrium
Chapter 11 ended with a warning: a cell at equilibrium is a dead cell. At equilibrium every reaction runs forwards and backwards equally, so nothing has a direction. Life needs direction: building proteins, copying DNA, moving muscles. It gets it by continually burning fuel to keep its chemistry pushed away from equilibrium, like a battery that's constantly recharged (chapter 13).
Here is the simplest version: molecules cycling between three forms, A → B → C → A. Without fuel, detailed balance holds (chapter 11): every step runs equally in both directions and there's no net circulation. Add fuel to one step and a steady current flows around the loop, a non-equilibrium steady state.
No net flow. At equilibrium each individual step is balanced, so going round one way is exactly as common as going round the other. That's detailed balance, and it's what makes equilibrium dead. Turn up the fuel and watch a net circulation appear, along with populations that no longer match the Boltzmann prediction.
The cell's main fuel is ATP. Turning ATP into ADP and phosphate releases about 30 kJ/mol under standard conditions. Inside a cell, where ATP is kept at hugely higher concentration than equilibrium would allow, it releases more like 50 kJ/mol. Couple that release to an uphill reaction (an enzyme makes them happen together, as one step) and the pair runs downhill. For example, making the amino acid glutamine costs about +14 kJ/mol on its own, and coupled to ATP the total is about −16 kJ/mol. Your body recycles roughly its own weight in ATP every day.
"ATP stores energy in a high-energy phosphate bond, released when the bond breaks." This is chapter 1's trap again. Breaking any bond costs energy. ATP hydrolysis releases free energy because the products (ADP, phosphate, and the water they interact with) are much more stable than the reactants: less repulsion between crowded negative charges, better stabilisation by water, more ways to arrange the electrons. And because the cell keeps ATP far from equilibrium, which is the part that really powers life. The energy is in the whole reaction and in the concentrations, not in one bond.
Chemistry and machine learning
You arrived here from ML, so here's where the two fields now meet.
- Protein structure. Predicting how a chain folds was a 50-year grand challenge. AlphaFold (DeepMind, 2020–21) learned to predict structures from sequence and evolutionary data with near-experimental accuracy for many proteins, and its developers shared the 2024 Nobel Prize in Chemistry (with David Baker, for protein design).
- Learned force fields. Machine-learned interatomic potentials learn the energy surface (chapter 6's PES) from quantum calculations. They then run chapter 1-style simulations at near-quantum accuracy for millions of atoms.
- Generative design and retrosynthesis. Models propose new molecules and materials, and plan routes to make them as searches over chapter 9's graph-rewrite rules.
Living chemistry is a non-equilibrium steady state: a Markov chain whose stationary distribution has a net probability current, because detailed balance is broken by an external drive. The same trick appears in non-reversible MCMC samplers ("lifting"), which add a persistent direction to the random walk and often mix faster than reversible ones. Protein folding is chapter 10's landscape in thousands of dimensions, shaped like a funnel by evolution so that the chain finds its native state quickly. AlphaFold sidesteps simulating that landscape by learning the map from sequence to structure directly.
Predicted is not made, and made is not useful. In 2023 Google DeepMind's GNoME reported 2.2 million predicted stable crystal structures; a 2024 critique argued that many were neither new nor practically useful. Generative models often propose molecules nobody knows how to synthesise. And AlphaFold predicts a static structure, not the motions and alternative shapes that often matter for function.
- Look
- Two small plastic bottles, warm (not hot) water, a spoonful of sugar, a packet of dry yeast, and two balloons.
- Wonder
- "Yeast is a tiny living thing. If we give it sugar to eat, what do you think will happen to the balloon?"
- Try
- Both bottles get warm water and yeast; only one gets sugar. Stretch a balloon over each and leave them somewhere warm for an hour. Which balloon grows?
- True sentence
- "Yeast are tiny living things. They eat the sugar and breathe out a gas, which blows up the balloon. Every living thing, including you, keeps taking in food and changing it into new stuff."
The gas is carbon dioxide; the yeast is fermenting the sugar, and the carbon atoms that were in the sugar end up in the balloon (atoms are conserved, chapter 9). The bottle without sugar is the control: it shows the sugar is what matters. It's the same reaction that makes bread rise.
Play more
- AlphaFold Protein Structure Database. Look up the predicted shape of almost any known protein.
- MolView. Draw a molecule and see it in 3-D.
- Nick Lane, The Vital Question and Transformer. Life as chemistry far from equilibrium.
- For the 6-year-old: bake bread together and watch the dough rise. Same yeast, same gas.