WEBVTT

00:00:01.004 --> 00:00:03.858
So how are we going to beat
this novel coronavirus?

00:00:04.091 --> 00:00:06.950
By using our best tools:

00:00:07.217 --> 00:00:09.585
our science and our technology.

00:00:09.802 --> 00:00:12.989
In my lab, we're using
the tools of artificial intelligence

00:00:12.989 --> 00:00:14.657
and synthetic biology

00:00:14.657 --> 00:00:18.285
to speed up the fight
against this pandemic.

00:00:18.285 --> 00:00:20.519
Our work was originally designed

00:00:20.519 --> 00:00:23.089
to tackle the antibiotic
resistance crisis.

00:00:23.089 --> 00:00:26.048
Our project seeks to harness
the power of machine learning

00:00:26.048 --> 00:00:29.517
to replenish our antibiotic arsenal

00:00:29.517 --> 00:00:33.120
and avoid a globally devastating
post-antibiotic era.

00:00:33.749 --> 00:00:37.075
Importantly, the same technology
can be used to search

00:00:37.075 --> 00:00:38.777
for antiviral compounds

00:00:38.777 --> 00:00:41.512
that could help us fight
the current pandemic.

NOTE Paragraph

00:00:42.310 --> 00:00:46.479
Machine learning is turning
the traditional model of drug discovery

00:00:46.479 --> 00:00:47.529
on its head.

00:00:47.529 --> 00:00:49.214
With this approach,

00:00:49.214 --> 00:00:51.465
instead of painstakingly testing
thousands of existing molecules

00:00:51.465 --> 00:00:53.217
one by one in a lab
for their effectiveness,

00:00:53.217 --> 00:00:56.203
we can train a computer

00:00:56.203 --> 00:01:00.430
to explore the exponentially larger space

00:01:00.430 --> 00:01:04.435
of essentially all possible molecules
that could be synthesized,

00:01:04.435 --> 00:01:10.014
and thus instead of looking
for a needle in a haystack,

00:01:10.014 --> 00:01:13.783
we can use the giant magnet
of computing power

00:01:13.783 --> 00:01:18.677
to find many needles
in multiple haystacks simultaneously.

NOTE Paragraph

00:01:18.677 --> 00:01:20.462
We've already had some early success.

00:01:20.462 --> 00:01:24.784
Recently, we used machine learning

00:01:24.784 --> 00:01:28.076
to discover new antibiotics
that can help us fight off

00:01:28.076 --> 00:01:32.547
the bacterial infections that can occur
alongside SARS-CoV-2 infections.

00:01:33.348 --> 00:01:35.784
Two months ago, TED's Audacious Project

00:01:35.784 --> 00:01:39.376
approved funding for us
to massively scale up our work

00:01:39.376 --> 00:01:44.415
with the goal of discovering
seven new classes of antibiotics

00:01:44.415 --> 00:01:47.384
against seven of the world's
deadly bacterial pathogens

00:01:47.992 --> 00:01:50.528
over the next seven years.

00:01:50.731 --> 00:01:52.428
For context,

00:01:52.428 --> 00:01:55.071
the number of new class of antibiotics

00:01:55.071 --> 00:01:58.290
that have been discovered
over the last three decades is zero.

NOTE Paragraph

00:01:58.404 --> 00:02:01.927
While the quest for new antibiotics
is for our medium-term future,

00:02:01.927 --> 00:02:06.435
the novel coronavirus
poses an immediate deadly threat,

00:02:06.602 --> 00:02:10.388
and I'm excited to share that we think
we can use the same technology

00:02:10.388 --> 00:02:13.548
to search for therapeutics
to fight this virus.

00:02:13.748 --> 00:02:15.549
So how are we going to do it?

00:02:15.549 --> 00:02:18.403
Well, we're creating
a compound training library,

00:02:18.403 --> 00:02:21.644
and with collaborators
applying these molecules

00:02:21.644 --> 00:02:23.957
to SARS-CoV-2-infected cells

00:02:23.957 --> 00:02:28.398
to see which of them exhibit
effective activity.

00:02:28.398 --> 00:02:31.538
These data will be use to train
a machine learning model

00:02:31.538 --> 00:02:35.709
that will be applied to a [?]
library of over a billion molecules

00:02:35.709 --> 00:02:40.570
to search for potential
novel antiviral compounds.

00:02:40.570 --> 00:02:43.289
We will synthesize and test
the top predictions

00:02:43.289 --> 00:02:46.275
and advance the most promising
candidates into the clinic.

NOTE Paragraph

00:02:46.508 --> 00:02:48.701
Sound too good to be true?

00:02:48.701 --> 00:02:49.869
Well, it shouldn't.

00:02:49.869 --> 00:02:53.276
The Antibiotics AI Project is founded
on our proof of concept research

00:02:53.276 --> 00:02:56.674
that led to the discovery
of a novel broad spectrum antibiotic

00:02:56.674 --> 00:02:57.859
called Halocin.

00:02:58.642 --> 00:03:01.478
Halocin has potent antibacterial activity

00:03:01.478 --> 00:03:05.615
against almost all antibiotic-resistant
bacterial pathogens,

00:03:05.801 --> 00:03:09.442
including untreatable
pan-resistant infections.

00:03:09.942 --> 00:03:12.211
Importantly, in contrast
to current antibiotics,

00:03:12.211 --> 00:03:16.215
the frequency at which bacteria
develop resistance against Halocin

00:03:16.215 --> 00:03:17.673
is remarkably low.

00:03:18.224 --> 00:03:23.178
We tested the ability of bacteria
to evolve resistance against Halocin

00:03:23.178 --> 00:03:25.299
as well as Cipro in the lab.

00:03:25.299 --> 00:03:27.131
In the case of Cipro,

00:03:27.131 --> 00:03:30.175
after just one day, we saw resistance.

00:03:30.391 --> 00:03:31.960
In the case of Halocin,

00:03:31.960 --> 00:03:34.646
after one day we didn't
see any resistance.

00:03:34.646 --> 00:03:38.039
Amazingly, after even 30 days,

00:03:38.039 --> 00:03:40.991
we didn't see any
resistance against Halocin.

NOTE Paragraph

00:03:41.242 --> 00:03:46.980
In this pilot project, we first tested
roughly 2,500 compounds against E. coli.

00:03:47.213 --> 00:03:50.273
This training set included
known antibiotics,

00:03:50.273 --> 00:03:52.176
such as Cipro and penicillin,

00:03:52.176 --> 00:03:55.127
as well as many drugs
that are not antibiotics.

00:03:55.294 --> 00:03:57.904
These data we used to train a model

00:03:57.904 --> 00:04:02.491
to learn molecular features
associated with antibacterial activity.

00:04:02.491 --> 00:04:05.542
We then applied this model
to a drug repurposing library

00:04:05.995 --> 00:04:08.187
consisting of several thousand molecules,

00:04:08.187 --> 00:04:10.456
and asked the model to identify molecules

00:04:10.456 --> 00:04:13.225
that are predicted
to have antibacterial properties

00:04:13.225 --> 00:04:15.910
but don't look like existing antibiotics.

00:04:16.194 --> 00:04:21.305
Interestingly, only one molecule
in that library fit these criteria,

00:04:21.506 --> 00:04:24.342
and that molecule
turned out to be Halocin.

00:04:24.691 --> 00:04:27.769
Given that Halocin does not look
like any existing antibiotic,

00:04:27.769 --> 00:04:31.822
it would have been impossible for a human,
including an antibiotic expert,

00:04:31.822 --> 00:04:34.724
to identify Halocin in this manner.

00:04:34.892 --> 00:04:37.545
Imagine now what we could do
with this technology

00:04:37.545 --> 00:04:40.053
against SARS-CoV-2.

NOTE Paragraph

00:04:40.053 --> 00:04:43.106
And that's not all.

00:04:43.106 --> 00:04:44.423
We're also using the tools
of synthetic biology,

00:04:44.423 --> 00:04:46.893
tinkering with DNA
and other cellular machinery,

00:04:46.893 --> 00:04:50.051
to serve human purposes
like combating COVID-19,

00:04:50.402 --> 00:04:54.312
and at [??] we are working
to develop a protective mask

00:04:54.411 --> 00:04:58.193
that can also serve
as a rapid diagnostic test.

00:04:58.414 --> 00:04:59.945
So how does that work?

00:04:59.945 --> 00:05:02.617
Well, we recently showed that you can take

00:05:02.617 --> 00:05:04.360
the cellular machinery
out of a living cell

00:05:04.360 --> 00:05:07.912
and freeze-dry it along with
RNA sensors onto paper

00:05:08.261 --> 00:05:12.916
in order to create low-cost
diagnostics for Ebola and Zika.

00:05:13.750 --> 00:05:18.429
The sensors are activated when
they're rehydrated by a patient sample

00:05:18.429 --> 00:05:21.563
that could consist of blood
or saliva, for example.

00:05:21.781 --> 00:05:25.040
It turns out this technology
is not limited to paper

00:05:25.040 --> 00:05:28.894
and can be applied
to other materials, including cloth.

00:05:28.894 --> 00:05:31.163
For the COVID-19 pandemic,
we're designing RNA sensors

00:05:31.163 --> 00:05:33.014
to detect the virus

00:05:33.014 --> 00:05:38.412
and freeze-drying these
along with the needed cellular machinery

00:05:38.412 --> 00:05:40.561
into the fabric of a face mask,

00:05:40.561 --> 00:05:42.851
where the simple act of breathing,

00:05:43.414 --> 00:05:47.725
along with the water vapor
that comes with it, can activate the test.

00:05:48.043 --> 00:05:51.496
Now, if the patient
is infected with SARS-CoV-2,

00:05:52.263 --> 00:05:56.350
the mask will produce a fluorescent signal
that can be detected by a simple,

00:05:56.350 --> 00:05:58.860
inexpensive handheld device.

00:05:58.860 --> 00:06:01.111
In one or two hours, a patient
could those be diagnosed

00:06:01.111 --> 00:06:06.339
safely, remotely and accurately.

NOTE Paragraph

00:06:07.006 --> 00:06:10.349
We're also using synthetic biology

00:06:10.349 --> 00:06:12.678
to design a candidate
vaccine for COVID-19.

00:06:12.912 --> 00:06:15.830
We are repurposing the BCG vaccine,

00:06:15.830 --> 00:06:18.975
which has been used against TB
for almost a century.

00:06:18.975 --> 00:06:20.459
It's a live attenuated vaccine,

00:06:20.459 --> 00:06:24.228
and we're engineering it
to express SARS-CoV-2 antigens,

00:06:24.777 --> 00:06:27.829
which should trigger the production
of protective antibodies

00:06:27.829 --> 00:06:29.649
by the immune system.

00:06:29.649 --> 00:06:33.544
Importantly, BCG is massively scalable
and has a safety profile

00:06:33.544 --> 00:06:37.012
that's among the best
of any reported vaccine.

NOTE Paragraph

00:06:38.096 --> 00:06:43.224
With the tools of synthetic biology
and artificial intelligence,

00:06:43.224 --> 00:06:46.611
we can win the fight
against this novel coronavirus.

00:06:46.844 --> 00:06:50.354
This work is in its very early stages,
but the promise is real.

00:06:50.354 --> 00:06:54.591
Science and technology
can give us an important advantage

00:06:54.591 --> 00:06:57.695
in the battle of human wits
versus the genes of superbugs,

00:06:57.695 --> 00:06:59.871
a battle we can win.

NOTE Paragraph

00:06:59.871 --> 00:07:01.655
Thank you.