人工智能的真相（2023）（2）

so it's almost like threading beads with all different shapes
各有不同
and properties onto a string.
这串珠子要叠成一个
And then that straight line has to fold up into a really
十分复杂的三维结构 就像这个模型
complicated three dimensional structure like this,
它叫溶菌酶 是眼泪中的一种蛋白质
which is lysozyme, a protein from our tears
能杀死进入我们眼里的细菌
that kills germs that fall into our eyes.
为什么了解这类蛋白质的结构
OK, so why is knowing the structure
很重要呢
like this of a protein so important?
这就像我们会拆解一个机器
Well, kind of like taking a machine apart to understand
来了解它的运作原理
the way it works.
但我们没法这样拆解蛋白质 因为太小了
We can't really do that with proteins because they're so small
它们是纳米级别的
and they're on the nano scale
而且存在于非常小的细胞里
and they're in our tiny, tiny cells.
所以我们很难搞清楚它们的形状
So we don't have an easy way to find out their shape.
但是如果知道了它们的形状 就能研究其功能
But if you know their shape, then you can work out what they do
它们如何连接 如何移♥动♥
and how they stick to each other and how they move.
跟钟表里的机械有点像
It's like clockwork, kind of.
所以如果我们知道了其形状 就能了解它们是如何
OK, so if you know their shape, you can understand how they can
帮助我们 或者伤害我们 对吗 -是的
help you or hurt you. Is that right? - Absolutely.
好了 在我们交谈的过程中
OK. Now while we've been talking,
想必观众们都发现了 在我们的头顶
it has not escaped our audience's notice that hanging above us
有一个看起来不小 有些危险 有些奇怪的装置
is a rather large, threatening, and strange contraption.
这是什么呢
So what is it?
这就是一个蛋白质
So apparently this is a protein,
当然要比我实验室里的大上5千万倍
and it's a lot bigger, 50 million times bigger than the ones in my lab.
所以这是一段氨基酸序列 对吗
So this is a sequence of amino acids, is that right?
这是一串氨基酸
The linear sequence of amino acids.
你肯定知道这是什么
I'm sure you can tell that's exactly what it is!
我们一眼就看出这是一串氨基酸了
We all knew immediately that this was a sequence of amino acids!
好了 我们马上会把它折叠起来
OK. Well, OK, we're going to fold this sequence in a moment.
在那之前 我想请大家猜猜
But before I do that, I want you to see if you can guess
它会是什么形状
what shape it's going to be.
有人愿意吗 前排的
Who wants to have a guess on the front?
你觉得它会是什么形状
What shape do you think it's going to be?
我觉得会是一串长方形
I think it might be a sequence of rectangles.
长方形 其他观众呢 那边的观众
A sequence of rectangles. Anybody else? You on this side?
我觉得会是五边形 -五边形 好的
I think it's going to be pentagons. - A pentagon? OK.
他们的猜测都很棒
Both really good guesses.
接下来我们就揭晓答案
But now we're going to find out.
请工作人员折叠蛋白质
OK, so, team, let's fold our protein!
好的
Ok!
这是皇家科学院的简写
It spells Royal Institution, and...
也是我名字的简写 -里夫卡·艾萨克森
My initials! - Rivka Isaacson!
这是我们的小设计
We planned that all along!
好的 里夫卡 是什么让这些氨基酸
OK. So, Rivka, what causes these amino acids to fold
以这个形状折叠起来了呢
in the way that they do?
举个例子 如果这里有一个正电荷
Well, so, for example, if you have a positive charge over here
这里有一个负电荷
and a negative charge over here,
那么它们就会互相靠近
then they want to be near each other
就像磁铁一样
in the same way that a magnet
正负极相吸
would attract another magnet.
氨基酸还有很多其他特质
And there are lots of other different properties
会让其处于三维结构中的
of the amino acids that want to be in particular places
特定位置
within the three dimensional structure.
这些都能帮助蛋白质找到它活性最高的位置
And that's what helps it find its most energetically favourable position.
所以研究蛋白质的结构
OK. So, finding out the structure of proteins
是你工作中重要的一部分
is a really important part of your job.
在人工智能出现前 你是怎么开展研究的
How did you do it before Ai?
有很多种方式
Well, there are lots of ways to do it,
但它们都价格昂贵 需要花很多年时间
but they're all very expensive and take years and years.
所以认真研究蛋白质的结构和折叠
And, so, getting to grips with protein structure, protein folding,
是一项非常重要的任务
is a really important problem.
里夫卡 感谢你的到来 并向我们讲解了
Rivka, thank you so much for coming and telling us about
这个及其困难的问题 -谢谢
such a difficult problem. Thank you very much. - Thank you.
那么人工智能是如何帮助我们的呢
So how can AI help?
让我们欢迎下一位科学家 他所在的团队
Let's welcome a scientist from the team responsible
正是有了人工智能的帮助 才得以取得
for one of the most important scientific advances
这项重大的科技进步
made possible by AI,
有请来自DeepMind的凯瑟琳·塔尼亚苏凡娜库
DeepMind's Kathryn Tunyasuvunakool!
凯瑟琳 欢迎来到圣诞讲座
Kathryn, welcome to the Christmas lectures.
跟我们解释一下什么是阿尔法折叠
So tell us, what is AlphaFold?
阿尔法折叠是我们开♥发♥的一项人工智能系统
So AlphaFold is an AI system that we developed for predicting
是用来预测蛋白质结构的
the structure of proteins.
它以氨基酸序列作为输入 然后...
And it takes as input that amino acid sequence and then...
就像我们在上面看见的那个巨大的东西一样
Like that big thing that we saw above us.
和那个巨型香肠串一样的排水管一摸一样
Exactly like that big sausage of drainpipes.
没错
Yeah.
然后它尝试输出折叠好的蛋白质
And then it tries to output the folded up protein.
你是怎么创造出它的
OK. So how did you create it?
阿尔法折叠是神经网络的一个例子
So AlphaFold is an example of a neural network,
我们用所有实验验证过的蛋白质结构训练它
and we trained it using all of the experimentally determined structures
就像里夫卡讲过的那种蛋白质
that Rivka talked about.
科学家孜孜不倦地验证了
So there are thousands of those that
几千种蛋白质的结构
have been painstakingly determined by scientists.
我们用这些蛋白质训练阿尔法折叠
We run AlphaFold on those training proteins
比较它的预测和真实的蛋白结构
and we compare its prediction against the true structure,
然后我们试着让它的预测更准确
and then we try and make it more accurate on those examples.
我想你可以向我们展示一下
So I think you can show us
里夫卡这样的科学家是如何
how AlphaFold is actually used
运用阿尔法折叠的
by scientists like Rivka?
当然
Absolutely.
阿尔法折叠可以在几分钟之内做出新的预测
So AlphaFold can produce a new prediction in a matter of minutes,
或者你可以在阿尔法折叠的数据库里
or you can look up a prediction in the AlphaFold
查找一个预测结构
protein structure database.
辅酶I
NAD
一旦你有了一个预测
And once you have a prediction,
它就可以被用来研究楚蛋白质的功能
it's useful for figuring out what the protein does,
如何相互作用 还能用来规划实验
how it interacts, and for planning experiments.
你可以看到
So you can see the structure of
我们预测的一种蛋白质的结构
one of our predicted proteins here,
我们还会提供置信度
and we actually provide confidence measures
来帮助科学家理解
to help the scientists understand
他们能信赖预测中的哪些部分
which parts of this prediction they should rely on.
所以我们看到的是一个真实的蛋白质吗
So this is a real protein we're looking at now?
没错 这是一个真实存在的蛋白质
This is a real protein, yes.
它确实很复杂 -的确
That is complex. - It really is.
它比刚刚那个RI复杂多了
It's much more complex than that RI example.
蓝色的区域
And when you look at the blue regions,
是模型认为准确度比较高的部分
those are the parts where the model is confident,
橙色的区域则是模型认为不那么可靠的部分
and the orange regions are parts where the model is less confident.
这有助于科学家利用这一结果预测
So that helps scientists use this prediction.
一旦♥科♥学家知道了这种结构
So once scientists know this kind of structure,
他们就可以专注于
that's then going to allow them to focus on the stuff
他们真正需要关注的部分 对吗
that they really need to focus on, right?
没错
Yeah, exactly.
结构生物学就是在解释这些东西
Structural biology is all about interpreting these things.
凯瑟琳 谢谢你来出席
OK, Kathryn, thank you so much for joining us
向我们讲解阿尔法折叠 谢谢
and telling us about AlphaFold. Thank you.
我邀请了德米斯·哈撒比斯来讲解
I asked Demis Hassabis to explain
阿尔法折叠对研究者的益处
the benefits of AlphaFold for researchers.
根据经验 通常一个博士学生
The rule of thumb is it normally takes one PhD student
需要用一整个博士学位的时间 也就是整整五年
their entire PhD, so up to five years,
用复杂的知识体系 比如X射线晶体学
just to find the structure of one protein
才能找到一种蛋白质的结构
using complicated systems like X-ray crystallography.
在过去几年里
And so we managed over the last couple of years
我们成功折叠了自然界中科学已知的
to fold all 200 million proteins known to science.
两亿种蛋白质
So, out there in nature.
只有其中的一小部分
And so only a tiny fraction of those
是之前通过实验找到折叠结构的
were known before experimentally.
如果你把博士学位的五年和两亿相乘
And if you times the five years of a PhD by 200 million,
那么阿尔法折叠就节省了十亿年的读博时间