(원문)

 

 

Automated machine conducts, assesses and learns from experiments with random reagents.

 

 

Reporter Adam Levy talks to chemist Lee Cronin about his team’s search1 for new chemical reactions. Read the research.

Subscribe to the Nature Podcast on iTunes or your favourite podcast app.

Head here for the Nature Podcast RSS feed.

 

 

TRANSCRIPT

Interviewer: Adam Levy

For chemist Lee Cronin chemistry is all about discovery. And despite methods and techniques that have developed leaps and bounds over the centuries, for Lee the process of discovery is still all about the journey.

Interviewee: Lee Cronin

It’s very difficult when you ask a chemist in the lab to go discover something. It’s a bit like asking someone to go in a boat now and find a new continent. Something invariably goes wrong and more interesting happens, and then we follow that up, and I think that is the point where conventionally a lot of discoveries are made today.

Interviewer: Adam Levy

But despite his love of the voyage, Lee thinks it’s time to change up the process. Not the aspect that relies on some magical combination of luck and intuition, but exactly how the search is carried out. Lee wants to automate the quest for new reactions. Machines already exist to automate certain tasks in chemistry, but they tend to follow specific programming and recipes rather than searching for new discoveries. There have also been attempts to aid discovery with machine learning, where a computer is trained with data and learns how reactions might operate. Lee wanted to make a machine that could learn and carry out tasks. I called him up to find out more.

Interviewee: Lee Cronin

So, what we’ve tried to do is use machine learning to classify where the outcome of a reaction – that’s when you mix two chemicals together – whether something has happened or not, and then you can use that as a basis to then, you know, navigate round if you like your new chemical space or sail your yacht around your unknown islands and map them.

Interviewer: Adam Levy

And your machine isn’t just learning, it’s doing, it’s doing experiments.

Interviewee: Lee Cronin

Yeah, we do three crucially important things. Number one, we start with an empty database, except just it knows the ingredients that we have at our disposal. Number two, it just randomly selects the ingredients, the chemicals to add together so it has no bias. And number three, it does this in real time so it actually decides what to do and then mixes the chemicals together and watches what happens.

Interviewer: Adam Levy

When we say ‘it’ in these sentences, we’re actually talking about a robot. Can you describe what this robot is, because it’s not like some kind of humanoid thing sitting at a chemistry lab table, right?

Interviewee: Lee Cronin

No, but actually it does things very similar to what a humanoid would do or a human being would do at the chemistry lab table. A chemist would typically mix chemicals together in a round-bottom flask, put them on the stirrer and heat it up, and that is what this robot does. You plug the chemicals into the robot, and it moves those chemicals as liquids in a solvent to the reactor, and then the reaction happens and the robot automatically takes a sample and moves it to a detector. And it’s basically like a human being using its eyes or your ears so it happens really quickly and seamlessly.

Interviewer: Adam Levy

So, is this really a closed system? Is this doing everything or is there still need for human input at some stage?

Interviewee: Lee Cronin

Oh yeah, the human’s crucial. This is not about replacing a human. So, what this robot does, it’s just a labour-saving tool. The robot is only as good as the chemist that’s trained it. Well, the robot would do the experiment and the chemist would tell the robot whether the outcome of the experiment was reactive or unreactive, so whether something happened or not. And so, the robot will then start to guess after a while and then the human will go yeah, you’re right, you’re right. And then there gets to a crucial point where it has done about 10% of the possible combinations, it is able to predict what will happen next and it just speeds up our ability to discover new reactions and new molecules. Human time is limited and so one of the things that the robot can do is basically do reactions the human doesn’t have time to do and would normally discard. It can do about 36 reactions a day, and a chemist would only typically do maybe 3 or 4 such reactions a day.

Interviewer: Adam Levy

We’ve described how the process works, but have you actually managed to find anything novel with this robot?

Interviewee: Lee Cronin

I’m pretty happy to say that I think I could convince maybe 9 out of 10 chemists that the robot had done some reactions where the outcome couldn’t be predicted beforehand, and that’s for me really exciting.

Interviewer: Adam Levy

Robots and automated machines are already used in chemistry, in industry. Just how different is the system that you’re using here from the kind of robots that might exist in other contexts in chemistry?

Interviewee: Lee Cronin

What’s different about our system is its integration and the fact it searches for reactivity. What we’re doing is actually quite unusual in that it basically is able to search without any target in mind, and then what we needed to make sure we were doing, not just having new sensors and not having targets, but actually having machine learning to actually correctly search those reagent or ingredient combinations.

Interviewer: Adam Levy

What actual applications will these differences be useful for?

Interviewee: Lee Cronin

Well, we’re really excited because we think in terms of discovery science, anything we need new molecules, so new drugs, new dyes, drug delivery systems, new materials. Now, there’s a problem when you’re discovering when you don’t know what you’re looking for. So, the next thing that we’re going to do is add another little sensor, but the sensor on to this will then have a desire and say right, we want to find, I don’t know, the bluest blue thing. So now, let’s not just look for new stuff but it has to be new and blue.

Interviewer: Adam Levy

And personally, what are you most excited about the opportunities that having this integrated robot system could open up?

Interviewee: Lee Cronin

Well, I’m hoping that what it will do is tell us more about the laws of chemistry, and allow us to discover molecules that we just wouldn’t have access to using our existing knowledge. I kind of liken it a bit like to writing poetry. Shakespeare was really interesting in writing poetry and verse because he invented new words. What I’m interested in this robot seeing it do, is able to invent new reactions. Then those reactions can be translated back to the normal chemistry language, and then the chemist is able to use those new reactions to make new molecules. For me that’s super exciting.

 

 

Controlling an organic synthesis robot with machine learning to search for new reactivity

 

 

Nature  559, 377–381 (2018)

 

 

 

Abstract

The discovery of chemical reactions is an inherently unpredictable and time-consuming process1. An attractive alternative is to predict reactivity, although relevant approaches, such as computer-aided reaction design, are still in their infancy2. Reaction prediction based on high-level quantum chemical methods is complex3, even for simple molecules. Although machine learning is powerful for data analysis4,5, its applications in chemistry are still being developed6. Inspired by strategies based on chemists’ intuition7, we propose that a reaction system controlled by a machine learning algorithm may be able to explore the space of chemical reactions quickly, especially if trained by an expert8. Here we present an organic synthesis robot that can perform chemical reactions and analysis faster than they can be performed manually, as well as predict the reactivity of possible reagent combinations after conducting a small number of experiments, thus effectively navigating chemical reaction space. By using machine learning for decision making, enabled by binary encoding of the chemical inputs, the reactions can be assessed in real time using nuclear magnetic resonance and infrared spectroscopy. The machine learning system was able to predict the reactivity of about 1,000 reaction combinations with accuracy greater than 80 per cent after considering the outcomes of slightly over 10 per cent of the dataset. This approach was also used to calculate the reactivity of published datasets. Further, by using real-time data from our robot, these predictions were followed up manually by a chemist, leading to the discovery of four reactions.

 

 

 

 

http://www.etnews.com/20180722000036

 

 

AI로 신물질 발견 앞당기는 ‘로봇 화학자’

 

 

글래스고대에서 개발한 로봇 화학자

 

 

 

의료, 신소재 등 영역에서 혁신을 가져올 새로운 물질을 개발하는 과정은 길고 어렵다. 우연한 행운이 없다면 수많은 실험을 거쳐 최적 특성을 가진 물질을 찾기까지 수년이 걸린다. 이 같은 어려움을 해소해 줄 ‘로봇 화학자’가 개발됐다. 인공지능(AI)으로 화학 반응 결과를 예측, 기간을 단축시킨다.

글래스고대 연구진은 최근 화학 반응 실험과 분자 발견을 가속화하는 로봇 과학자를 공개했다. 이 로봇은 AI 방법론인 기계학습(머신러닝)을 사용해 화학 반응을 예측할 수 있다. 이를 위해 직접 실험을 통해 수집된 데이터를 활용했다.

이 로봇은 18가지 화학물질을 이용해 1000가지 반응이 나오는 실험에서 80% 정확도로 연구에 적합한 반응만 추려내기 위해 100번의 실험만 필요했다. 연구진은 이 실험으로 네 가지 적합한 반응을 발견했으며 이 가운데 한 건은 상위 1%에 드는 독특한 반응이었다고 설명했다.

이번 실험 결과가 엄청난 성공처럼 보이지 않을 수도 있다. 그러나 이 기술은 화학자가 신물질을 발견하는 속도를 획기적으로 개선할 수 있다. 화학자는 AI가 예측한 성공 가능성이 높은 소수 반응에 초점을 맞춰 연구, 탐색 과정을 단축할 수 있다. 알고리즘도 지속적인 데이터 축적과 학습을 통해 정교해질 것으로 전망된다.

미국 정보기술(IT) 매체 엔가젯은 “과학자는 로봇 화학자를 이용해 일상 실험이 아니라 까다로운 부분에만 집중하면 된다”면서 “새 치료법, 새 배터리 소재, 더욱 강도 높은 물질 발견 등 다양한 영역에서 연구 기간을 크게 줄일 수 있을 것”이라고 말했다.

 

 

 

 

 

 

 

Leave a Reply

Your email address will not be published. Required fields are marked *