Nuclear-war head: assassination in Iran

Hello and welcome to The Intelligence on Economist Radio. I'm your host, Jason Palmer. Every weekday we provide a fresh perspective on the events shaping your world. Nearly every process in biology involves different proteins, with each one's shape defining its purpose. But those shapes are incredibly hard to work out. So a speedy new artificial intelligence approach could solve all kinds of problems in disease and drug design. And about a month ago, as COVID-19 cases rose sharply, a number of European countries went back into lockdowns. We look into streams of mobility data and ask why this round of lighter-touch lockdowns was so surprisingly effective. But first: The funeral of Iran's top nuclear scientist underlined his significance to the country. Yesterday's ceremony for Mohsen Fakhrizadeh was attended by top-level officials at the Defense Ministry in Tehran. A military detail A detail then carried his flag-draped coffin through the streets to a cemetery in the north of the capital. Mr. Fakhrizadeh was shot dead while driving last Friday. He'd been at the heart of Iran's clandestine nuclear ambitions for decades. Iran has blamed Israel for the murder, a charge that formally Israel refutes. Iran's president, Hassan Rouhani, has promised retaliation. But any retribution might be tempered by hopes of renewed diplomacy with America.

There are a number of countries in the Middle East, and namely Israel and Saudi Arabia, that don't want to see Iran brought in from the cold.

Roger McShane is our Middle East editor.

And one way to look at the killing of Mohsen Fakhrizadeh is as an attempt to sort of scorch the earth before Joe Biden enters office and tries to restart diplomacy with Iran.

And what do we know so far about Mr. Fakhrizadeh's killing?

Well, Iran first said that gunmen carried out the attack in a roadside ambush. Now it says it was carried out with a machine gun operated by remote control. The question is who was operating the remote? And according to the secretary of the Supreme National Security Council in Iran, Ali Shamkhani, that would be Israel.

And what do you make of that accusation then?

Well, I mean, Israel has a history of targeting Iranian nuclear scientists. It's killed a number of them over the years, although not many recently. And it's killed them in rather sort of brazen and shocking attacks that involve sort of motorcyclists speeding past moving cars and attaching magnetic bombs to them. Benjamin Netanyahu, Israel's prime minister, has been somewhat coy about this attack. But if you remember in 2018, He pointed to Fakhrizadeh by name as a key player in Iran's nuclear program.

A key part of the plan was to form new organizations to continue the work. This is how Dr.

Mohsen Fakhrizadeh, head of Project Amad, put it.

Remember that name, Fakhrizadeh.

And so in terms of the health of that program, how important is this assassination?

Fakhrizadeh had been quite important. I mean, he led a series of these innocuous-sounding institutions such as the physics research center. But he was always thought to be the driving force behind Iran's secretive nuclear weapons program. He was basically Iran's institutional know-how. He was the most important repository of the country's nuclear weapons knowledge. But this attack would have been more effective a decade ago. For starters, as of 2019, America believed that Iran had not restarted its weapons program. And if it wanted to restart that program today, it could do so without Fakhrizadeh because it has a lot more competent nuclear scientists than it did a decade ago. So, you know, that's all to say that the attack was perhaps as much symbolic as strategic.

And how likely is Iran to take action now against whoever it thinks is responsible?

Yeah, that's the big question. I mean, its leaders have vowed revenge, but they always do that. And so far they haven't done anything. In general, Iran has been relatively quiet this year, even though this is just the latest in a string of attacks inside the country that really shown up its weakness. Remember that the year started with the assassination of its most revered general by America, to which it responded by lobbing some missiles at American bases in Iraq but killed no one. So it may be waiting to see if Joe Biden pursues a different approach than Donald Trump, who pulled America out of the nuclear deal that Iran negotiated with world powers in 2015 and who has heaped sanctions upon Iran, cutting it off from the world economy and causing it Tremendous pain.

Well, how do things stand now with what the Trump administration has tried to do?

Yeah, well, in recent years Iran had stopped complying with parts of the nuclear deal because Trump sanctions ensured that it wasn't getting much out of it. But Biden has said that if Iran moves back into compliance with the deal, then America would rejoin it. And he's been quite critical of Trump's whole approach to Iran.

President Trump has no strategy here, it seems to me. He has no endgame. And here's the hardest truth of all. His constant mistakes and poor decision-making have left us, the United States, with a severely limited slate of options for how to move forward. And most of those options are now bad.

Now, as much as the killing raises tensions between America and Iran, it will probably complicate matters. But one could also argue that from the Biden team's perspective, the Trump administration's belligerence on the way out the door isn't the worst thing in the world, actually, as much as it sort of highlights what will inevitably be a change in tone. And as much as it gives the new administration sort of leverage to negotiate a better deal.

But rejoining the Iran nuclear deal, the JCPOA, is not just a matter of putting a signature on paper, right? It's more complicated than that.

Yeah, exactly. If Joe Biden were just to jump right back into the nuclear deal, And that would entail giving up a lot of the leverage I spoke about. In the words of Jake Sullivan, the incoming National Security Advisor, Iran would also have to be prepared to advance good faith negotiations on follow-on agreements. Those agreements would deal with things that weren't contained in the first agreement. Things like Iran's missile program, its use of proxies to cause chaos and destruction in places like Yemen, in Syria, in Iraq. And also, it would deal with the sunset clauses in the first agreement. The first agreement, much of it expires in 5 years, and the rest of it expires in 10. So the administration has given itself some room to maneuver. But it will be interesting to see what it means when it says that Iran must enter these good faith negotiations on follow-on agreements. Does that mean that Biden is first going to join the nuclear deal and then he can negotiate the agreements? Or does it mean that all of this has to happen at the same time?

I mean, there are a lot of ifs in there. As things stand now, do you think that there's a chance that Iran could return to the table, that a push more for diplomacy is possibly on offer here?

Yeah, I mean, we sort of forget that as much as there's been big changes in America, there have also been big changes in Iran. The pragmatists in Iran who negotiated the original nuclear deal have been completely discredited by the failure of that deal to deliver economic benefits. Hardliners are now ascendant and attacks like the one on Fakhrizadeh really enforce their worldview. Iran is going to hold its own presidential election in June, and most expect a hardliner to win. Before then, it has a choice. It can go with the hardliners and retaliate for the killing of Fakhrizadeh, or it can wait for Biden and see if it can make a fresh start with the new administration.

Roger, thank you very much for joining us.

My pleasure.

The science of proteins is at once a simple bit of biology and one of the most devilishly complex. The proteins involved in just about every biological process on Earth are all long strings of just 20 or so building blocks called amino acids. But just like every book in English is made up of the 26 building blocks of the alphabet, that simple starting set leads to dazzling variety. For reading, all that matters is the order in which the letters appear. For proteins, it gets quickly more complicated, because how a protein functions—and whether it does—has everything to do with how it folds up once it's assembled inside a cell. Determining those precise shapes is therefore a central issue for all manner of biological questions, from understanding diseases to designing drugs. Painstaking work over decades, using X-rays and lasers and microscopes, has revealed some of those structures, and supercomputers have helped matters along. But results revealed this week suggest a shortcut in the form of artificial intelligence. Cracking long-standing protein problems with ease.

So yesterday, a group of researchers from DeepMind, which is an AI lab owned by Google, they presented some results at a scientific conference on using computers to predict the final shape of a protein from nothing more than a list of its amino acid components.

Tim Cross is our technology editor.

And if it stands up, it's potentially quite a big deal because This is something that's hard to do, so it could mean big changes for pretty much the entire discipline of biology.

So why is that such a hard thing to do experimentally? I mean, the various sorts of microscopes out there don't cut it?

It's not that they don't cut it. We know how to do this in principle. So you could crystallize proteins, you can put them in tiny little crystals, shine a whole load of X-rays at them, watch how the X-rays scatter, and then kind of work backwards from that to infer where all the amino acids in a protein must go. It's just that it takes a long time and it's a lot of work, and for some proteins it's not always clear that you've got the right answer. There's a lot of interpretation involved in doing all this. So there's a sort of interesting mismatch in biology where, because of cheap gene sequencing, we know of 180 million proteins where we can tell you what the chemical formula is. But actually doing the experiments necessary to find the structure, that's complicated enough that we only have about 170,000 of those. So there's several orders of magnitude difference here, and that simply arises because the experiments are time-consuming and a bit awkward.

But it seems like the sort of thing that computers should be able to tackle pretty straightforwardly.

It seems like it, but it turns out to be much more complicated than that in practice. So when you really drill down to it, this is basically a problem in applied quantum mechanics. So a protein is just a very long molecule. Molecules are made of atoms. There are all kinds of weird and wonderful atomic-scale forces that act between those atoms and within them and so on. And the sum total of all those forces basically determines how this protein, which starts life as a long unfolded— you can think of it like beads on a string. Those forces basically determine how it folds itself up into the correct shape and also why it gets the correct shape instead of one of the zillions of different other shapes it could take almost every time. The trouble is, we know in computer science that it's just not feasible because it would take you longer than the lifetime of the universe to do the number of sums that you need to do. It just runs away with you for any kind of complicated molecule. So what people have been trying to do for the last 50 years is to try and come up with conceptual shortcuts that would take this problem and make it tractable for computers, but also still make the result reasonably accurate.

So how does the artificial intelligence approach to this get past those limitations?

What they've done is to use an AI technique called machine learning, which is what powers facial recognition and voice assistants. And there is a computer game out there called Foldit, which is a sort of scientific computer game where people are given models of proteins and they're invited to sort of play around with them and see if they can fold them into the desired shapes. 'Cause once you've done the folding, you can work out if the shape is good or not. The difficult bit is just working out what the folds should be. And there's this interesting result where it turns out that even amateur people with no biology experience can get pretty good at playing with Foldit. And so the thinking is that there obviously are shortcuts that you can take, 'cause these people presumably aren't doing trillions of quantum physics calculations in their head. DeepMind will say that if you talk to people who play Foldit, they'll just say, "Oh, well, you know, I do this by intuition," or, "I make moves 'cause it feels right." And what that tells you is that there are kind of rules of thumb there, you just have to work out what they are, 'cause people can't tell you explicitly. From there, you take a whole pile of data. In this case, it was about 170,000 protein structures and the amino acid sequences that are associated with them. You feed them to a machine, a whole bunch of complex computer science stuff happens in the background, and the result is you end up with a machine that's internalized a lot of those rules and can apply them to new proteins that it's never seen before.

So you say that if this indeed stands up, this would be transformative for biology. I mean, the ability to simply compute what a folded-up protein looks like, where does that get us? What does that allow us to do?

I think it allows us to do all kinds of things, and the biologists I've talked to about this are all pretty excited. It could teach us about disease. There are diseases out there like Alzheimer's which people strongly suspect are at least partly caused by misfolded proteins, when something goes wrong in the folding and you end up with a protein not only that doesn't work but is actively harmful. DeepMind have tried experimentally, and we don't have any good data on this, but they say that they let this system loose on the coronavirus and it generated protein structures that pretty closely match what we found from experiments. So maybe the next time we have a global pandemic like this, we'll be able to react that much faster. You could maybe use it to speed up the development of drugs by getting a better sense of what it is they're doing or how the proteins that they're targeting work. You could screen them for unwanted interactions or maybe find new uses for existing drugs. You might even be able to use this stuff to design synthetic proteins that don't exist in nature but that you want because they can speed up this or that industrial process. I think there's potentially all kinds of applications.

Tim, thank you very much for joining us. Thanks, Jason. 4 weeks ago, England went into another lockdown following similar tightening in France, Germany, and Italy.

All our friends around the world are grappling with the same question of how to keep people safe without retreating into a winter of hibernation.

Today is the last day of those restrictions. Then a new regional tiered system will go into force. The British government says the November restrictions helped dramatically reduce COVID-19 case numbers and pushed the so-called R-naught below 1.

In England, in the week before Before the peak, the number of cases grew by 11%. But in the last week, cases have dropped by 30%, almost a third.

Our data team has been digging into just why the lockdown was so effective.

Facing a surge in infections in October and November, 11 countries in Europe reimposed national stay-at-home orders.

Sondre Solstad is a senior data journalist at The Economist.

We find that they have been limiting people's movement less than the lockdowns in the spring, but still been effective in reducing new infections.

How do we know, though, how much people have been moving around?

Cell phone data from Google confirms that these lockdowns had a smaller impact on people's lives. So we saw in March that when lockdowns were imposed, the average number of trips fell to 35% of the level in January. So people moved around 65% less. During these lockdowns, it dropped to only 68%. The smaller effect of lockdowns this time could be caused by people not adhering to them as strictly. But there is also another explanation, which was that they were designed with more care and in a situation where more information was available about the disease. So they were more flexible and allowed more exceptions.

And so in light of those mobility data, then how have the lockdowns affected the actual spreading of the disease?

So they have been quite effective in the sense of bringing the epidemics under control. So you see the average R in the week before lockdowns were imposed was about 1.1 for these countries, whereas in the week after they were imposed, it was dropped to 0.9. That small change makes a big difference. Over just 4 weeks, it would mean new infections falling by 21% rather than rising by 36%.

And why is that?

Well, us and the data team have crunched the numbers on 334 local authorities in England and Wales, roughly akin to American counties, to see how lockdowns led to a reduction in R.

And so what do you see in those data?

What we found was that the effect of lockdowns was primarily through two channels. So the first was reducing trips to workplaces, and then to a greater extent, reducing trips to leisure activities such as restaurants, pubs, and retail shopping. But the biggest effect was actually on reducing trips to other local authorities. So it led to people staying more within their local area, and this was the way the lockdowns had the greatest effect on reducing R.

Having looked through all of these data, does it lead to prescriptions on how lockdowns could be better designed?

I think it does. So there's been a lot of talk about how reducing transmission by reducing trips to the pub, trips to restaurants, and so on is important. And I think that is important. But what the unique data we got and crunched for the UK shows is that Equally, or perhaps even more important, is reducing transmission between areas. So reducing the transmission that goes from county to county could be one smart way to design future lockdowns.

Sondre, thank you very much for your time.

Always a pleasure to be here, Jason.

That's all for this episode of The Intelligence. If you like us, give us a rating on Apple Podcasts, and you can subscribe to The Economist at economist.com/intelligenceoffer. The link is in the show notes. See you back here tomorrow.