AI for scienceMiddleLesson 2536 min read

AI for drug and materials discovery

AI can screen huge numbers of candidate molecules and even design new ones. But every candidate still faces a long road of lab tests and, for medicines, clinical trials.

✦ Ask the tutor🧪 Lab · 30 min📝 Lesson test
Lesson in motion

In 60 seconds

AI for drug and materials discovery

AI can screen huge numbers of candidate molecules and even design new ones. But every candidate still faces a long road of lab tests and, for medicines, clinical trials.

1/6
In simple words
Imagine you must find one key that opens a lock, and there is a mountain of keys. A smart helper can sort the mountain and give you a small bag of keys that look most likely to fit. You still have to try each key in the lock. And some keys that fit might break the lock!
Finding a new medicine or a new material is a search problem. The number of possible molecules is enormous, far more than anyone could make and test. AI helps by making the search smarter. It does not remove the need to test.

Two ways AI helps

ApproachWhat it doesExample use
ScreeningA model predicts properties for a big list of existing or easy-to-make molecules and ranks them.Ranking which chemicals might bind to a target protein, or might be stable as a battery material.
Generative designA model proposes new molecules or materials that are not on any list, aiming for chosen properties.Suggesting a new molecule shape that might fit a protein pocket and still be easy to make.
Both approaches give candidates. A candidate is a guess with a reason behind it. Many candidates that look good on a computer fail in the lab: they cannot be made, they fall apart, or they do not work as predicted.

The long road for a medicine

  1. 1

    Pick a target

    Often a protein involved in a disease. Structure predictions can help here.
  2. 2

    Find candidates

    Screening and generative design suggest molecules. This is where AI helps most today.
  3. 3

    Make and test in the lab

    Chemists make the molecules. Tests check if they work on the target and on cells.
  4. 4

    Improve

    Change the molecule to work better and cause fewer side effects. Repeat many times.
  5. 5

    Safety tests

    Careful studies before any human takes the medicine.
  6. 6

    Clinical trials

    Tests in people, in stages, with ethics approval and informed consent.
  7. 7

    Regulator review

    In India, the Central Drugs Standard Control Organisation (CDSCO) reviews new medicines. Other countries have their own regulators.
Most candidates drop out somewhere on this road. AI may help the early steps go faster or pick better candidates. The later steps, especially clinical trials, still take a long time because safety in real people must be tested carefully.

Materials: batteries, solar cells and more

The same idea works for materials. Scientists want better battery materials, solar cells, catalysts that clean up chemical processes, and materials for cheaper water filters. Models can predict whether a material is likely to be stable and what properties it might have. Then the promising ones must be made and measured.
A material that is stable on a computer may be very hard to make, or may be too expensive, or may use rare or toxic elements. Real-world use adds many questions a model does not answer.

Reading the headlines

Watch out
Headlines like "AI discovers new drug" are common. Often the real news is that a model suggested a candidate that did well in early tests. That is interesting, but it is far from a medicine you can buy. Never change any treatment because of a news story. Talk to a doctor.
  • What stage is it? Computer prediction, lab test, animal study, or human trial?
  • What exactly did AI do? Rank candidates, design a molecule, or analyse data?
  • What was compared? Better than what? Was there a fair baseline?
  • Who checked it? Is there a peer-reviewed paper or only a press release?
  • What is still unknown? Good articles say what has not been shown yet.
Hype
  • "AI cures disease X"
  • No stage mentioned
  • Only the company's own press release
Honest reporting
  • "A model suggested a candidate that worked in early lab tests"
  • Clear stage and next steps
  • A paper others can read and check
Danger
Generative models for molecules can in principle propose harmful chemicals as easily as helpful ones. Responsible labs control access, review risky requests and follow chemical and biosafety rules. Never try to make chemicals based on AI suggestions outside a supervised lab.
Try this. Find one recent news story about AI and a new medicine or material. Using the five questions above, decide what stage the work is really at.

Watch and read more

Lab

A marked-up science press release with a short, fair summary of what it really shows.

~30 min

The problem

Find a press release or news article about AI helping discover a drug or material. Good places: a university news page, a science magazine, or a national newspaper's science section. If you cannot find one, use the made-up release below.

"Researchers at a university lab announced that their new AI system has discovered a revolutionary compound that could transform the treatment of fever. The AI screened a large library of molecules in just days. The top compound reduced a fever marker in cells grown in the lab. 'This could change everything,' said the team leader."
  1. Underline every word that sounds like hype (for example "revolutionary", "could change everything").
  2. Answer the five questions from the lesson: stage, what AI did, what was compared, who checked it, what is unknown.
  3. Rewrite the first two sentences so they are accurate and calm.
  4. Write one question you would ask the researchers.

You are done when

Hard questions

Try to answer before you reveal. If you can answer these, you understood the lesson.

Q1The release says the AI screened molecules "in days". Why is speed alone not the right measure of success?Reveal

Fast screening is only useful if it picks better candidates than other methods. The fair question is: of the candidates the AI picked, how many worked in the lab, compared with a baseline method? Speed at the start also saves little if the long later stages, like safety tests and trials, stay the same.

Q2Why might a promising result in cells fail in animals or people?Reveal

Cells in a dish are a much simpler system. In a whole body, the molecule must be absorbed, reach the right place, not be broken down too fast, and not harm other organs. It can also interact with other molecules and medicines. Each of these is a new chance to fail, which is why the road is long.

Please sign in to continue.

Questions people ask

Has AI made new medicines that people take today?

AI tools are now used in many drug discovery projects, and some AI-assisted candidates have entered clinical trials. It is hard to say simply "AI made this medicine", because many people and methods are involved. Check the latest reliable sources, and be careful with headlines.

Why can't we skip trials if the model is very good?

The human body is far more complex than any model. A molecule can work on its target and still harm the liver, or interact with other medicines. Only careful testing in people, with safety rules and consent, can show this.

What is a baseline and why does it matter?

A baseline is a simple method you compare against, for example a random pick or an older, standard method. If an AI method is not clearly better than a fair baseline, the result is not impressive, however fancy the model is.

Can a student work in this area?

Yes. Chemistry, biology, computing and statistics all help. Public data sets and open-source tools let students practise on real molecules. Start by learning the science basics well; the AI part is easier with that foundation.

Lesson test

5 questions. Get 3 right (60%) to pass and complete this lesson.

Sign in to take the test and save your progress