ai drug discovery

An orchestra of a hundred musicians plays without a conductor. Every instrument is tuned perfectly, but one violin goes slightly off, and the harmony falls apart.

A stain is found at a crime scene. The expert takes a swab, and a few hours later the instrument outputs a list of proteins that made up the fluid. Previously, a rapid test could be wrong if multiple fluids were present: blood mixed with saliva, and the strip gave a false result.

An interpreter gets distracted by every rustle in the hall, and a one-hour meeting stretches into two. Similarly, molecular foundation models learn the language of molecules from vast corpora, but each new conversation requires fine-tuning. Monroe changes the rules: it doesn't learn a new language but instantly finds common meaning from a few examples.

Hunting for a new material is like cooking without a recipe: you try, try again, and often throw things out. Previously, scientists manually flipped through papers and tested compositions one by one. Now there's a "robot chef" that reads the literature itself, mixes components, and checks what came out.

When sugar goes into tea, it dissolves. But why doesn't it just fall back out? It's all about chemical potential — a quantity that shows how 'favorable' it is for a substance to be in a given place and state. For a pure substance, it's easy to calculate, but for a mixture of several components, it's a real puzzle.

A protein slides along the DNA strand like a key along a keyring, searching for its lock. One wrong turn and a gene stays switched off. To know where a protein will grip firmly, you need to measure the strength of that grip — the binding energy.

Each morning, a navigator plots a route, bypassing traffic jams and roadblocks. Similarly, before a radiation therapy session, a new AI agent recalculates the irradiation path.

At intersections without traffic lights, people negotiate with glances. A self-driving car can't do that: it has to simultaneously decide whether to go or yield and plan its trajectory. Previously, these tasks were separated, making maneuvers jerky and dangerous. Scientists have combined them using virtual force fields enveloping each car.

A robot swarm is like a cloud of fireflies: each moves on instinct to wherever is freer, and chaos turns into order. For a long time, it was a mystery how coherence emerges without communication or commands. The answer came with the creation of the Response Map — a tool that, like an X-ray, illuminates the hidden logic of the machine brain.

Building a light chip is like a game where you have to guide a beam through a maze of tiny mirrors and beam splitters. Each element acts like a railroad switch: it either lets light pass straight through or directs it to an adjacent track. The goal is to produce a specific light pattern at the output while using as few 'switches' as possible. Usually, engineers spend years finding the shortest route for each task.

DFTB is a high-speed train in the world of quantum chemistry: hundreds of times faster than classical DFT, yet still on electronic rails. But its engine — iterative self-consistency of charges — sometimes resembles an orchestra where each instrument starts from an arbitrary note.

Inside each of us, invisible clocks are ticking. They control the release of cortisol — the hormone that helps us wake up in the morning, cope with anxiety, and keep our immune system on its toes. In adrenal insufficiency, these clocks break down: the body can't produce the right amount of cortisol on its own.

Think of a massive cookbook: each recipe is a protein, and the finished dish is its job in the cell. To whip up a complex treat, you often need to mix several ingredients and let them 'set' into the right shape. Until now, scientist-chefs could only guess how someone else's recipes would fold.

Imagine an audition for a symphony orchestra, where out of hundreds of hopefuls, only those whose instruments sound pure and withstand the furious crescendo remain. In the world of materials, computers have learned to generate 'atomic scores'—thousands of virtual crystals that promise to become the foundation for quantum chips or superconductors.

Gold doesn't lie on the surface: it's hidden in tons of barren rock. To extract a gram of precious metal, prospectors sift through cubic meters of dirt.

Creating a material with desired properties is an art that long relied on intuition and luck. Nature gives no hints: dozens of elements in the periodic table, and the number of their combinations tends to infinity. Previously, materials scientists resembled alchemists, trying everything in a row, or relied on statistical models that lost power beyond the known.

A solid electrolyte is not just a crystal lattice: it’s a bustling metropolis for ions. Each act of conduction resembles a trip through the city: there are wide highways with low energy barriers, gridlocks in bottlenecks, and tricky detours where an ion can get stuck.

An engineer tunes an antenna: pull one segment, the frequency drifts; change the angle, the gain drops. Developing therapeutic peptides (short protein chains) works much the same way.

Perovskite crystals harvest sunlight almost as well as silicon, but cost pennies. Their trouble is an inborn fragility: from warmth and ordinary water in the air, they “get sick.” Inside, disorder builds up, and the shiny plate turns cloudy, losing its abilities within a couple of days.

Computational studies of material behavior have long suffered from two extremes: either fast but inaccurate models, or accurate but slow physical calculations. Modern programs, trained on data, have nearly resolved this dilemma — now the bottleneck is not computing power, but human routine: setting up simulations and checking results. Paimon works like an autonomous chef.

Creating an alloy from several metals is like cooking a soup from mysterious ingredients: guessing the final texture without trial and error is nearly impossible. Conventional models often get it wrong when there are more than three different kinds of atoms.

A lab sends a robot a list of a thousand candidate molecules for a new drug. The predictor robot, trained on old data, gives each one a 'fit or no fit' score. But when the list comes from an unfamiliar area, a different virus or a rare tumor, the robot starts to get confused. Its confidence stays high, while actual hits drop to zero.

Picking out a single faulty microchip from a million, without any clue what the fault looks like, is a classic needle-in-a-haystack problem. A classical computer would test each chip one by one.

Imagine a huge garage with hundreds of parking spots. You're driving a tiny car, a future drug molecule, and you want to find the one free bay where it fits perfectly. But these bays aren't concrete; they're woven from DNA or RNA chains, and no one has posted the parking rules.

In a car, the steering, brakes, and accelerator gradually fail. At first, you can still swerve around a pothole, but each year the maneuver takes more effort. At some point, even a perfect road becomes impassable — the car no longer obeys the driver. That's exactly how the authors describe aging: not as the accumulation of rust or wear and tear, but as the loss of "controllability" of the body.

A drop of blood from a vein falls under three beams, and each one drags its own clues into the light. The infrared beam makes molecules vibrate — from that trembling, you learn how much sugar and fat is in the sample. A green laser scatters off proteins and lipids, shifting color — revealing the remodeling of cell membranes.

Inside every cell, life is bustling: tiny protein machines tirelessly haul cargo, read genes, and fix errors. But their work is far from perfect: they often 'spin their wheels,' take backward steps, or get stuck in futile cycles. It's like a city traffic jam where cars dart around aimlessly, burning fuel.

A laser beam is split into dozens of thin filaments, passes through frosted glass, and reconverges to a single point. This is the workbench where the program took a seat. It received no instructions — only a general goal: figure out how light changes when passing through a scattering medium, and find something new.

The light bulb turns on only in the room where you need it. That's roughly how photoactivatable drugs work: they 'switch on' only where light is directed, leaving the rest of the body untouched. This is especially important in cancer treatment: conventional chemotherapy hits all dividing cells indiscriminately, causing severe side effects.

Searching for a new material is like looking for a needle in a haystack, but here the haystack is millions of possible atomic combinations. In the past, scientists spent years trying out options in the lab, relying on intuition and luck. Now, artificial intelligence is stepping in, learning to 'understand' the language of crystals.

In a lab where crystals are studied, a technician stares at a screen covered in bright spots and slowly rotates the sample, trying to get the spots to form a symmetric pattern. It's like tuning an old radio: you turn the knob until the static gives way to a clear signal. Now, a machine can turn that knob. Scientists have created a program that learns to align crystals without any knowledge of how those spots work.

The best AI assistant for scientists scored worse on this exam than a C student on a pop quiz. Just 21% correct answers, and that's the ceiling for the four strongest models. The problems for the new test were written not by programmers but by graduate students: people who deal with real science every day. The test is called COMPOSITE-STEM.

A translator who has never been in a meeting translates literally, and loses half the meaning. They don't know that the boss's nod means "do it like last time," and the phrase "well, we'll see" is a polite refusal. That's how a language model that writes code from a scientific paper works: it can read the equations, but not the unspoken agreements.

Creating a new material, like a tough alloy, requires thousands of computer experiments. Usually, a scientist manually writes instructions and hunts for errors. AutoMOOSE is a robotic kitchen: say 'I want a light and strong metal,' and five software chefs whip up a recipe, launch the synthesis, and adjust if something goes wrong.

A seven-metal alloy is like a team where every player is pulling the blanket their own way. Atoms of iron, cobalt, nickel, aluminum, titanium, tantalum, and vanadium can either fall into a strict pattern or scatter into random positions. This determines whether the material will be as strong as steel or as light as aluminum.

A physicist analyzing particle collisions typically spends weeks on dull work: cleaning noise from data, estimating how many false signals will slip through, double-checking every number. Now that routine can be handed to a program.

Capsids are protein boxes that viruses use to carry their genetic material. In gene therapy, these boxes serve as couriers: you put a desired gene inside, and the outside retains the ability to enter cells.

Inside each tiny particle that makes up the cathode of a sodium-ion battery, sodium is unevenly distributed — like sugar clumping in tea. This reduces the battery's capacity and lifespan. To see this hidden landscape, scientists used X-ray microscopy and trained artificial intelligence to interpret the data.

Science is like a city where residents pave the streets themselves. Where paths are already worn, new houses go up, while untouched vacant lots go unnoticed — not because building there is impossible, but because no one ever looks.

Previously, generative models for drug discovery worked like an artist painting without choosing a canvas. They learned from existing molecules and simply produced similar ones, without considering how the space of possible compounds is organized. SpaceGFN changes the approach: now chemical space is a construction set where the user specifies the parts and the assembly rules.