Three Hundred Million Structures

JH Research.

The CEO's office.

I was watching the Nano people paint on the big TV, the scene looking for all the world like an art-school entrance class. A hundred or so of them were gathered, pencils flying over paper.

"That's something else."

They were rendering proteins in exacting detail. Scratch, scratch. Each of them had a magnifier wedged against one eye, studying a protein and transferring what they saw onto the page.

"Wow. They're good."

One hundred people selected from a hundred thousand. Over ten thousand had applied, which meant these were the ones who could simply look at something and put it on paper.

"These are the real golden hands."

And it wasn't just humans. Dwarves were mixed in among them, drawing too. Of course they were. A race born artisans and destined to die artisans lived up to the name. At minimum, Leonardo da Vinci class.

"These ones are just printers."

If the human artists looked like golden hands, the dwarves looked like printers. "Not even the same league" was a phrase invented for moments like this. Of course, some of the humans outdrew even the dwarves.

The mages.

Where the dwarves spat out stills like printers, the mages produced video. Proteins weren't fixed. They swayed and shifted, free to move within a limited range—and with more than one joint in play, the motion got viciously complex. Capturing all of that in animation? Magic made it possible.

"Stare long enough and even I start to get it."

Proteins. A protein was a chain of amino-acid lumps linked end to end—exactly like a string of sausages. Twenty-two kinds of amino acids. With residues like A, R, and K, you got sequences like M—A—R—K… lined up one after another into that long sausage chain.

"I heard it takes a university lab a full year to work out a single structure."

One Nano person took about half a day on one protein. A rough sketch would have been fast, but they produced dozens of sheets per protein. Flip those drawings in sequence and you had an animation. With a hundred people drawing, a thousand protein structures would be fully resolved in three dimensions in five days.

A job that would have taken a thousand years in a single Earth lab.

They just looked at the proteins and drew. Looked with their eyes. Skilled hands going scratch, scratch.

Easy, right?

For them it was simple. Earth's researchers were losing their minds trying to figure out things they couldn't see.

There was one problem, though.

"At this point, getting protein samples is the harder part."

For the Nano people to draw, they needed proteins to look at. ‍‍‌‌⁠​​⁠‌​​‍​​‍​‌​⁠‍‌‍⁠⁠‌⁠‌‍‌‍‌‌‌‍⁠​‌​‌​‌‍‌‌‌⁠‍​​‌‍⁠‌‌⁠⁠Samples. I had called around everywhere and sourced them. Common proteins were easy enough, but once we passed a thousand kinds, new samples got scarce.

"Just acquiring the samples ran a billion won."

I could have shopped slowly and cheaper, but the Nano people drew so fast I paid premiums to keep the pipeline full. Money always bought time. That way, the structures of over a thousand protein samples were fully solved and handed off to Nano AI.

"So Nano AI has trained on more than a thousand protein structures now?"

Seeing once beats hearing a hundred times. What method could possibly beat showing an AI the exact structures, drawn by eye? Nano AI should be starting to understand protein structure any day now.

Beep-beep.

The office intercom rang.

"Hello?"

Something in Gong's voice sounded urgent.

"What is it?"

"Strange how?"

"We did."

"What?"

Ah.

Of course it had—the Nano people had drawn the structures and fully trained the AI on a thousand of them. Our brilliant Nano AI had been handed a thousand perfect examples and simply decided it now knew everything about proteins.

Getting samples had already been getting hard, so an AI that claimed mastery of proteins was a relief. The rarer something was, the more the price climbed exponentially. Protein samples were no different.

"I'll come down."

I switched off the TV and went down to the AI room. Gong greeted me with ‍‍‌‌⁠​​⁠‌​​‍​​‍​‌​⁠‍‌‍⁠⁠‌⁠‌‍‌‍‌‌‌‍⁠​‌​‌​‌‍‌‌‌⁠‍​​‌‍⁠‌‌⁠⁠a face lit by pure excitement.

"CEO, this is wrong. There's no way protein-structure training should be finished already… but it says complete."

A little explanation ‍‍‌‌⁠​​⁠‌​​‍​​‍​‌​⁠‍‌‍⁠⁠‌⁠‌‍‌‍‌‌‌‍⁠​‌​‌​‌‍‌‌‌⁠‍​​‌‍⁠‌‌⁠⁠seemed in order.

"Ah—that would be because I fed it some data."

"Even so—"

"I put in a thousand."

"A thousand what?"

"A thousand fully solved protein structures as training data."

"There was a set like that?"

"Yeah. I dug around and got hold of them. So right now, our Nano AI has perfectly understood a thousand past exam questions."

Gong's eyes went bright.

"I knew it. You pulled something, CEO."

"Something like that. So what's next?"

"Since it claims complete understanding through a thousand solved protein-structure problems, the next step is to feed it the amino-acid sequence of a new protein and have it analyze the structure."

"Are there a lot of proteins where only the amino-acid sequence is known?"

"On the order of several million."

"That many?"

"It's just the order of the amino acids listed out. The AI now has to figure out position, angle, folding, and shape on its own."

"Let's try it."

"Yes, sir."

Click-clack. Gong typed in the amino-acid sequence of a new protein.

A—R—R—K—A…

A simple sentence of three hundred letters. As data, maybe a kilobyte if that. From that alone, Nano AI now had to construct the three-dimensional shape of the sausage chain those amino acids would form.

Whirrrr.

The progress bar filled, slow and steady.

"Done."

"How do we look at it?"

"One second."

Gong fiddled for a moment.

"There. You can view it as a 3D simulation now."

I took the mouse and worked the simulation. Those three hundred simple letters I had just typed in had become a complex three-dimensional protein structure.

"It really is complicated-looking."

"It is. Team Leader—this isn't just a single line, is it? If it's one line, why is it this tangled?"

"A ball of yarn is one line too. It's twisted and knotted."

"Right. This part is coiled up like a spring."

"This is why scientists suffer."

A protein was a single chain of amino acids strung together. But that chain didn't sit in a neat row. It folded, twisted, spun, zigzagged.

"I'm more amazed that any scientists managed to work out even a few of these structures than that the AI figured them out."

"Same. It's like reconstructing a ball of yarn from its shadow alone. A year per structure isn't empty talk."

"Team Leader Gong, let's feed it more. This is fun."

"Understood. I'll dump in every known amino-acid sequence we have."

"That's all available?"

"Yes. Amino-acid sequences are plain text. Even a few million lines, the whole set is under two gigs."

Gong loaded every known amino-acid sequence. Our data center wasn't ordinary—two gigs vanished without even a noticeable pause.

"Two gigs is basically the same as hitting Enter."

"Exactly. Why do you think I never go home?"

"You should still go home sometimes."

"You're solving something that was supposed to take twenty years in a few days. It's too fun to leave."

While Gong and I talked, Nano AI was grinding hard.

The count of completed structures climbed without pause.

"Whoa. Instant."

"That's how AI works. The first teaching is hard. Once it knows how, scaling the numbers is nothing."

It looked that way. Ten thousand. Fifty thousand. Already over a hundred thousand protein structures were being calculated and spat out. I grabbed the mouse and clicked through finished structures. Click a name with its amino-acid sequence and the 3D form appeared. Flick, flick. I could spin the protein any way I wanted with the mouse.

"Nice."

Then, as I moved the cursor, text appeared on the model.

"Team Leader Gong, what's this?"

"Ah—so that feature is already in. It marks which part of the protein reacts with which substance."

"So you're saying a hydroxide ion can click right into this spot?"

"Exactly."

"Now I get why knowing protein structure is like a map for drug development."

"Seeing it with your eyes is completely intuitive."

"This protein binds hydroxide, so you use something with a hydroxide ion. That one has a hydrogen-ion site, so you use hydrogen. It's right there on the screen."

"That's right."

"Team Leader Gong—even laypeople like us can read this at a glance. For specialists, this is basically a map hack, isn't it?"

"Having a map hack for drug development doesn't sound like a bad thing."

"No, it doesn't."

Whirr, whirr. While we talked, Nano AI finished every structure.

"Man, that's fast."

"Haha. A few days ago we were saying twenty years. I still can't believe it."

While Gong and I were laughing, Nano AI asked a strange question.

What was that supposed to mean?

"Team Leader Gong, what is this saying?"

"Predictable proteins… maybe it wants to build structures for amino-acid sequences beyond what we entered?"

"So it's offering to tell us the structures of unknown proteins that haven't even been discovered yet?"

"Sounds like it."

"Want to try?"

"Yes, sir."

Whirr-whirr. Nano AI started analyzing again.

Ten minutes later.

Gong and I locked eyes.

"Three hundred million."

"CEO… what did we just do?"

"So we solved the 3D structures of the few million proteins that only had known sequences—and on top of that, we generated three hundred million structures for unknown proteins whose sequences aren't even known yet."

Gulp. Gong swallowed hard.

"CEO, what are you going to do?"

"About what?"

"No matter how you look at it, this is a map hack. Sell it to a pharma company and you'd make an insane amount of money."

Money that would make an ordinary person's eyes roll back in their head.

"CEO, if you sell this, it's a jackpot."

"And if I don't?"

"If you release it instead of selling, it would drag humanity's medicine forward overnight."

"Yeah. They called it a fifty-year unsolved problem. If the body and drugs are a lock and key, we just got the full anatomy of the lock. All that's left is cutting the right keys. Oh—right. Doing this was supposed to be Nobel material, wasn't it?"

"It was. For the drug-development field this is basically publishing a map hack. A Nobel would be the least they could do."

"Then should we release it?"

"Hiiik—if you sell this, you'd make a fortune—"

"How about we each take three deep breaths first?"

"Yes. Good. In… in… in…"

Gong and I breathed.

"Alright. Let's restate it. We built a map hack for drug development. Correct?"

"Correct."

"Release it, Nobel Prize. Sell it, a fortune. Or I start a pharma company myself and make an even bigger fortune a little later. Right?"

"Right."

I thought for a second. The choice came fast.

"Team Leader—there's one thing, though."

I dropped my gaze to my legs.

"As you can see, my legs are like this. And I'm already making an absurd amount of money."

I had already achieved financial freedom. Not just me—all of JH. Saying the company couldn't spend its money fast enough was about right. I hadn't gone into AI to make even more cash. If money alone had been the goal, I would have rented out the data center or leased the AI, or just scaled LPU sales.

Once you had enough money, you wanted to spend it raising the odds of treating yourself and your family.

"Team Leader Gong. The map hack for drug development. Let's release it clean."

"Yes, sir!"

Click-clack. Gong started hammering the keyboard.

"This is why I can't go home."

"Why? Does the release take that long?"

"No. I want to see the reaction after it goes live."

Hmm…

But couldn't he watch that from home?