AI for infertility, and scar-free healing
This month we hear about an artificial intelligence (AI) breakthrough for infertility, how ketamine can mimic some of the decision-making difficulties seen in schizophrenia, a new device to observe and document mosquito feeding behaviour, the key to scar-free wound healing, and how open is open access publishing at the moment?
In this episode
00:41 - AI improves IVF success rates
AI improves IVF success rates
Hadi Shafiee, Harvard University
Infertility is a major problem in developed countries worldwide. Just between 2017 and 2018, in America it fell by 2%. This means that, increasingly, couples are resorting to assisted conception methods, like IVF, to help them to conceive. This usually involves collecting eggs from the woman and sperm from the man, fertilising one with the other, allowing the resulting embryos to develop for a short period in the dish, and then choosing the best looking embryos to put back into the uterus to - hopefully - trigger a pregnancy. But this is a very subjective technique; the success rates are still relatively low at around 30%, and even among those couples who succeed, they usually need to go through several heart-rending and exhausting rounds of this process. So Hadi Shafiee, from Harvard, wondered whether an AI system could help, as he told Chris Smith…
Hadi - One of the most important factors is the ability to select the highest quality embryo and the current methods to do that are either manual, highly subjective, expensive, and time-consuming. So we wanted to tackle this problem by developing a machine learning-based, fully automated approach, to help embryologists to pick the best quality embryo.
Chris - Normally, an embryologist would look down a microscope at a selection of developing embryos, and they would say, "I reckon that one looks like the best prospect, followed by that one". And those are the ones they would cherry-pick to put back into the woman, aren't they, you're saying, can we use computers to help us do it even better?
Hadi - You got it right. Actually, we performed a very simple but very important experiment before this publication that we had at eLife. We recruited several embryologists with different backgrounds, you know, from two years to eight years of experience, and asked them, by providing them like hundreds of images of embryos with known outcomes, basically asked them to perform two simple clinical decisions that they normally do in their practice. And after they made their decision, we just rotated the images and asked them again to perform the same process on the same images. And there was huge variation between the decisions that they made.
Chris - So in other words, looking at the same embryo twice, they didn't make the same decision all the time, every time, showing there's enormous variation when they do it by eye?
Hadi - That is correct. So one application of using AI is to literally help them to have to at least reduce the subjectivity of the process. So basically how the system works is that we provide thousands of embryo images with the known outcomes. And then the machine intuitively kind of figures out what kind of features it had to emphasize in order to make the right decision.
Chris - If you do a trial and compare its performance against what an embryologist could achieve, how much better is the AI system?
Hadi - Very, very good question. So in fact, we did that study by recruiting 15 different embryologists from five different fertility centres across the country. And we provided images of embryos with really high quality. When you look at them morphologically they look amazing and we asked them to make their decision based on its implantation outcome. And we asked the same question from the machine. And we found that the machine actually did a better job, statistically, compared to the embryologist. 75% versus 60-something% on average.
Chris - That's quite a big difference. Do you actually know what it's looking for?
Hadi - At this stage it is literally a black box. We literally don't know what kind of features actually the machine was looking at.
Chris - Does the machine then make any surprising decisions? Does it ever disagree with you where it chooses an embryo and you'd think well in a million years, I'd never put that in?
Hadi - Very good question. Yes. In some cases it does surprise. You know, we literally saw embryos that actually, if you show it to an embryologist, the embryologist wouldn't transfer it. But then the machine picked that and then it ended up with a successful pregnancy. In some of the cases, it actually does surprise you in an interesting way, rather than in a negative way.
Chris - So if this were applied to your average IVF clinic, where rates are currently running at about a 30% success rate, what sort of success rate would this return? All other things being equal?
Hadi - In order to get a very confident answer, you literally need to perform a prospective clinical trial where you have a randomised study. But when you look at the retrospective data we have done, our machines showed that we were doing better even compared to MGH fertility centre, which is literally among the top 10% facility centres across the country.
06:01 - Ketamine mimics schizophrenia
Ketamine mimics schizophrenia
Sean Cavanagh, University College London
People with schizophrenia experience a range of symptoms, one being difficulty weighing up different pieces of information in order to make an informed decision. Now, UCL’s Sean Cavanagh has used both a computer model and tests on monkeys to show that low doses of the drug ketamine damps down the activity of NMDA receptors in the synapses connecting nerve cells in the brain to produce the same decision-making deficit seen in patients, as he told Chris Smith…
Sean - The way that we kind of approached this to try to relate the problems at the synapse with the behavioral symptoms, was that we built a computerized neural network, basically a computer simulation made up of nerve cells. It was made up of synapses and these synapses had the NMDA receptors that we were interested in. We then used this neural network to test our ideas. So we said, what happens with this neural network if we interfere with the NMDA receptors? So this gave us specific predictions about how decision-making would change and how neural activity would change. And then we could subsequently test these in an experiment.
Chris - You used the computer to come up with some testable ideas basically, that you could then go to a real brain and ask if we do this, does this happen as the computer suggests?
Sean - Yeah, exactly. That's what we did.
Chris - And so what did you do to actually test it?
Sean - The research subjects that we used for our experiment were monkeys. And the way in which we tested it was by using the drug ketamine, because ketamine is known to block the NMDA receptors. And ketamine is also an excellent experimental model of schizophrenia. As we know when ketamine is given to healthy human volunteers it temporarily reproduces many of the symptoms of schizophrenia, such as the hallucinations, the delusions and the cognitive symptoms.
Chris - What decisions were you asking the monkeys to make?
Sean - The types of decisions that we were interested in were those which involve combining multiple pieces of information. For instance, when we're deciding where to go on holiday, we have to combine lots of different attributes to make that decision. We might be thinking, for instance, of how excited we are about a possible destination, how good the weather will be, what the cost might be, and probably unfortunately at the moment, wherever we'd have to quarantine when we return. So you have to combine all of these different features. Obviously we didn't get the monkeys to decide where to go on holiday, but what we did get them to decide was on a sequence of images, they were showing some different bars. They had to add up the heights of all these different bars, combine lots of pieces of information and decide whether a series of bars on the left or right of a computer screen were higher.
Chris - And your intervention is that you present exactly the same task, but with or without the presence of ketamine. So you can see what role the NMDA receptor, that's going to be hit by the ketamine, is playing in helping them to resolve various stimuli to make a decision.
Sean - Exactly. So what we were trying to achieve was to study how the monkeys make their decisions normally, and then through using the ketamine, because we know it simulates the symptoms of schizophrenia and it blocks NMDA receptors, that's basically giving us a window or an opportunity to what we think might be going on in the brains of people with schizophrenia.
Chris - Before you tell us what happened to the monkeys when you did this, if you ask a person who does and a person who doesn't have schizophrenia to do these sorts of tasks, what different outcomes do you get between the two groups?
Sean - We know from previous studies they have impairments in decision-making. They struggle with combining different pieces of information. They make their decisions off only a small proportion of the information they’re actually shown. And so we come back to the example of deciding about a holiday, we had how excited you were, we had the weather, and we had the quarantine rules. If a person with schizophrenia was making that decision, they may only use the first attribute that they consider. They wouldn't consider any of the other attributes.
Chris - And the monkeys given the ketamine, did they show impairments in doing this simple decision making task based on bringing information together in this way?
Sean - Yeah, they did. So it's important to remember that we used a very small dose of ketamine. So the monkeys were still very engaged. They still were performing a task. They're still enjoying the task, but they just became slightly less accurate in their decision-making. So when they went not administered with ketamine, they were getting it right maybe 85% of the time. When they were administered with ketamine, their accuracy went down to maybe 70%. But it wasn't just they were getting worse at the decisions, it was they got specifically worse in a way that we had predicted with our computer simulation.
Chris - Do you know why that is? And similarly, why a person with schizophrenia would struggle to integrate all these different bits of information together? Is it that they just can't hold all the information in working memory at once and therefore make valid comparisons? Or is there something else going on to explain why their decision-making is falling in this way?
Sean - I think that the NMDA receptor, which is part of the synapse, is important for communicating between different nerve cells. And once this communication breaks down the neurocircuits that they're not able to combine information as well.
Chris - Does this point, Sean, to any ways in which we might present information in a way that is more meaningful for people who have this sort of problem?
Sean - It's important to stress again that we didn't test any people with schizophrenia in this study, although that's something we will be doing in future. But you are correct to say that for using these kind of sophisticated behavioral tests, we've worked out ways in which they may struggle to combine information. So it's also correct that we could also use these behavioral techniques to try to change the way in which these people are able to combine information. So that would be one approach. The other thing that is important to remember is that now we've identified some things which we think may be going wrong in the neurocircuits, the next step can be to work out how we can fix this. So we could think that if we had a therapy or some sort of drug, which could restore the activity of these NMDA receptors, then possibly this may be able to improve the symptoms of patients.
13:20 - The biteOscope: studying mosquito feeding
The biteOscope: studying mosquito feeding
Felix Hol, Institut Pasteur
There are thousands of species of mosquito, all with their own patterns of feeding behaviour, things that they are repelled by or attracted to, and diseases that they can carry. Studying them consistently, and realistically, is therefore a challenge. But now Felix Hol, from the Institut Pasteur in Paris, has come up with a gadget that he calls “the biteoscope”. As he told Chris Smith, this is an apparatus that’s home to a hoard of hungry mosquitoes and a patch of synthetic skin containing a blood-like fluid; there’s a camera below the skin patch hooked up to a computer to document how the insects behave…
Felix - We now just use basically the thing that's most easily thought of as synthetic skin. So basically just a fake human mimic. To us it just looks like a piece of plastic, but we treat the piece of plastic in such a way that mosquitoes find it attractive enough to find it, land it and bite it. We, for instance, heat the substrate close to body temperature, which is a very strong guiding cue for mosquitoes. We also can spray in CO2, which is normally in our exhaled breath and that activates the mosquitoes to start searching for a host. So we can guide the mosquitoes towards this little device and there they land on a membrane, and the mosquitoes actually need to pierce through a surface in order to be able to feed. They insert their mouth parts and here they find an artificial meal that is transparent. It has basically essential qualities of blood that make the mosquito think it is like blood, and they actually engorge as if they would be drinking blood.
Chris - Where's the camera then? Is that underneath that transparent membrane so that you're basically seeing the mosquito from underneath as it lands and then tries to feed?
Felix - Yes, this is exactly right. So the mosquitoes are just contained in the cage and in the floor of that cage, we have the bite substrate up. We can just image through the bite substrate while they're feeding on the substrate
Chris - And that data, that's all going into a computer, does this mean that you can basically use image analysis on it so that rather than you painstakingly picking apart what the mosquitoes are doing, timing how long it does each step, you can get a computer to do that? So you can do this potentially at enormous scale, cause you could study loads of mosquitoes without actually having to do anything yourself?
Felix - Exactly yes. So we developed a lot of algorithms, a lot of custom software that picks out the mosquitoes in those images - where it was, how long it was walking on the substrate, when it starts biting the substrate. And from those images, we can also obtain things like how much does their abdomen swell. So of course their belly swells while they're drinking and from how much the belly swells, we can actually calculate whether or not it's drinking and whether it gets fully engorged or not. You can track exactly where all the legs are going, where the mouth parts are going, where the head is. And from that, we can basically calculate the relative positions of the body parts.
Chris - Do you think the mosquitoes are actually fooled by this system though? Do they actually really regard this as skin and are they behaving in a way that you think is exactly the same as they would were it your arm in there for real?
Felix - So I do think that indeed they're fooled. They binge on this artificial meal, they drink a lot of it. They swell three times their volume. So in that sense, they're clearly fooled at the moment. We don't really know what the relative attraction of our substrate is compared to a human arm. I mean, that would be an interesting follow up experiment to do at some point where basically we present a real arm to the mosquitoes, and we present our substrate and we look at whether they prefer the arm over our substrate, or maybe the other way around
Chris - Now you've got this working, have you actually begun to gather data from it yet that are highlighting things that we previously, because of the constraints, the way we've historically tried to study mosquito feeding behavior, had overlooked?
Felix - Yeah. So there's one example in the study that Anopheles coluzzi that is a mosquito species that is an important malaria vector in Africa. People had known for a long time that it is rebelled by the mosquito repellent DEET. However, the mode of repellency was not entirely understood. And in this study we actually show that the mosquitoes senses it with their legs. We could coat a small portion of the bite substrate with this insect repellent. And when they land just outside of their coated area, we see that they just walk around and have their normal behavior. Whereas when they land in the coated area, they immediately take off right after their first touch. So from this we learned that actually the mosquitoes are likely tasting this repellent with their legs and are using that information to leave. And I think in the future there's many different things that we can do in this setup. For instance, one thing that I'm very excited about is to compare the behavior of mosquitoes that are infected with a pathogen, for instance, dengue virus or malaria parasites, versus the behaviour of mosquitoes that are not infected with anything in the literature. There are certain reasons to think that infected mosquitoes actually may behave differently. Maybe they're more avid or less avid to feed. Possibly the infected mosquitoes will take a smaller blood meal and therefore maybe actually will need to bite more often and therefore actually infect more people. These things are fairly difficult to measure because it's pretty difficult to do experiments on infected mosquitoes. I really think that with the BiteOscope, we will actually have a really cool angle on that and see if they actually behave differently.
18:57 - Healing wounds without scars
Healing wounds without scars
Ryan Driskell, Washington State University
Quickly healing a wound is a key part of our skin’s defence against bacteria and other microbes that could cause infection. The price we pay for that speed, though, is often a scar. But can we make skin regenerate without scarring, or even reproduce complex mini-organs like hair follicles that are also often lost to injury? It must be possible, because developing foetuses can do it, and as Eva Higginbotham heard from Washington State University’s Ryan Driskell, he thinks he’s found a gene that might be the key to the process...
Ryan - So there's a lot of research that was done by Michael Longacre and his mentor, and they were doing these life saving in utero surgeries. And what he found was when those babies were born, they didn't have a scar, even though he had done surgery in utero on those fetuses. And that was a remarkable discovery. So one of the things we wanted to do is identify not only a molecular factor, but also the cell types that are lost during skin maturation that would convey this ability of embryonic or neonatal skin to be able to regenerate. And the first step that we did was we started to do single cell RNA sequencing. So single cell RNA sequencing is a new technique that's been developed so that you can sequence, as much as possible, all of the genes that are expressed within a cell. And then from there, you can sequence thousands of cells within a tissue to identify cell populations. And using that technique on say, comparing all of the genes within a cell of over 5 to 10,000 cells in young skin, and then comparing that same assay in old skin, you could find new cell types as well as the genes that are defining those cell types. And that's what we did.
Eva - I see. So you found new cell types that were involved in really effective scarless wound healing and young skin.
Ryan - I would say we rediscovered them and we verified that they were there and that those cell types, in that cellular state, disappear in older skin. And we defined that cellular state by at least one gene - there are other genes that we found as well, but the gene that we decided to choose was LEF-1 and they're there in the young skin and they're not there once the skin matures.
Eva - So what is LEF-1 doing?
Ryan - Right. So LEF-1 is a transcription factor. A transcription factor is a protein that binds to DNA, and many transcription factors bind to DNA to control the regulation of genes to be expressed or sometimes to be inhibited. LEF-1 has been identified to be able to modulate signals from the outside of the cell called Wnts that come from other cell types or through its own signaling. And it's a part of a signaling pathway that's important during development and for some reason is somewhat turned off as the skin ages.
Eva - So once you found LEF-1, what did you do next?
Ryan - So some of the cells are still there, they just look more aged, right? They don't have the same phenotype, the state, the molecular state as measured by single cell RNA sequencing, is not there. So we hypothesized that if we could turn on LEF-1 in those older cells and keep it on, then it's possible that that transcription factor could sit on the DNA and be ready for a signal such as those that happen during wound healing. So we collaborated and found them a mouse that allowed us to control the expression of that transcription factor through transgenic technology, and it allowed us to express LEF-1 in those older fibroblasts. And when we did that, we created all different kinds of wounds at different time points. And, except for one time point, all regenerated with this enhanced stability that we hadn't seen before, with the erector pillys so that those wound hair follicles can even stand up if the mouse got cold.
Eva - Now that we know the importance of this factor, of LEF-1, what can we do with that information? Do you think one day we might have, you know, a cream that has Lef1 in it that we can rub on our skin if we get wounded?
Ryan - Right. I don't think you could put LEF-1 in a cream because transcription factors are notorious to be able to put inside of a cell. So the idea would be, and what we're trying to do now, is to understand what LEF-1 is doing in those special cell types, right? What are those downstream targets that LEF-1 is activating in response to a wound. And then I think we would actually want to try to understand this in humans.
24:02 - How open is open access?
How open is open access?
Chun-Kai Karl Huang, Curtin University
eLife is a digital pioneer. Since its launch the journal has published exclusively online and is open access at source, which means anyone, anywhere can freely access the research papers. Of course, many journals do not operate this way and instead place their content behind paywalls, depriving those with no subscription. Funders in many countries have introduced policies in the meantime that the work they’re supporting should be freely available in an open access format. So, how is the research community doing? Chris Smith heard from Chun-Kai Karl Huang, who's been trying to find out...
Chun-Kai Karl - My name is Chun-Kai Karl Haung, and I'm from Curtin University. We started with the project wanting to know how institutions and universities are performing in terms of making their research output freely accessible to people. Through that process, we're able to pick up several things, including the effect of policies and several other findings that should have implications to how we proceed in going forward in terms of making knowledge more accessible.
Chris - What was the period that you were considering?
Chun-Kai Karl - The results that we have includes data that goes back 50 years and the most recent 30 years are reported in the article itself.
Chris - And how have things changed?
Chun-Kai Karl - In the last decade there has been a lot of focus on making knowledge accessible. Lots of research has gone into research showing that making knowledge accessible to people is good for making more knowledge. And there's been a lot of initiatives in recent decades pushing this agenda forward. And we see that in the data. There has been a very low level of open access, maybe going 20, 30 years back, but in recent years, there has been a lot of push for that to happen. And we do see the open access level going up exponentially in recent years.
Chris - What's been the main driver of that shift towards more open access?
Chun-Kai Karl - There's a lot of open access initiatives. People have been pushing and researchers are working very hard in terms of advocating for open access. Not only in terms of showing research in terms of advantages of open access, but also implementing policies at regional levels, at country levels as well, able to push the agenda forward.
Chris - So how did you do this study? What did you actually measure?
Chun-Kai Karl - Our data framework measures the percentage of output for each university in which they make their research freely accessible. We do this for more than a thousand universities globally. What we found was that there are different policies and different resources that are having an underlying effect on how universities are performing. In terms of Australian universities, the current status is that universities in Australia lag behind universities in Latin America in terms of making research accessible through journals, but also lag behind universities, for example, in the UK, in terms of making publications available through repositories. A lot of that is to do with the strength of the policy and infrastructure that's provided to support making things open access.
Chris - There's an old saying "what gets measured gets done". And if you're saying that there are different performers in different geographies, say Australia versus UK, is that because there are policies that also include monitoring in one place that are not present or so rigorously enforced in another place, and as a result, less research is translating into the open access space in one particular territory than another?
Chun-Kai Karl - Yes, that's correct. Our data shows that places where there's more likely to be monitoring and possibly sanctioning of making research open access, places where the strength of the policy is stronger, they tend to be more open access. However, this is just one particular string of the story, because we suspect that it's a lot of combinations of different things.
Chris - What's going on in Latin America then, that means that there's such a high representation of research output from there, which is going open access? Is it because it is facilitated and incentivized, and made cheap for institutions there, whereas a first world institution like an Australian university or UK institution, there's a financial deterrent. Do we know?
Chun-Kai Karl - For Latin American universities, we think that a lot of the open access levels of the universities are being supported by a project like the CLO project, which has been providing infrastructure to hosting open access journals, which has made the whole cost of open access much more affordable. We think that is why Latin American universities have such high levels in gold open access, providing office access through the journals. Of course, there's the issue of funding pushing through open access in a short term. But there's also examples after a few years there's saturation.
Chris - What do you mean by saturation?
Chun-Kai Karl - So we do see the effects of the funding, but that effect seems to slow down as time goes by and if there's no extra funding.
Chris - What would be your recommendations then if there was a sort of a list of things that the best performers could teach to the worst performers, what would they be?
Chun-Kai Karl - Well, the first thing we observed that really worked well and is cost effective is providing infrastructure, like in the case of the CLO project in Latin America, that seems to have a much longer term effect. So that's the first point I would recommend to have some kind of infrastructure support for universities. Secondly, we have seen cases, for example, in the UK, even though there has been extra funding provided around 2012 for open access, a lot of the law of the push for open access in recent years has come through university library repositories. So that's possibly another route. We might want to support universities in going forward.