0:00:00.917,0:00:03.825
So how are we going to beat[br]this novel coronavirus?

0:00:04.317,0:00:06.948
By using our best tools:

0:00:06.972,0:00:09.011
our science and our technology.

0:00:09.594,0:00:12.726
In my lab, we're using[br]the tools of artificial intelligence

0:00:12.750,0:00:14.329
and synthetic biology

0:00:14.353,0:00:17.413
to speed up the fight[br]against this pandemic.

0:00:18.078,0:00:19.941
Our work was originally designed

0:00:19.965,0:00:22.818
to tackle the antibiotic[br]resistance crisis.

0:00:22.842,0:00:27.531
Our project seeks to harness[br]the power of machine learning

0:00:27.555,0:00:29.401
to replenish our antibiotic arsenal

0:00:29.425,0:00:33.263
and avoid a globally devastating[br]postantibiotic era.

0:00:33.685,0:00:36.505
Importantly, the same[br]technology can be used

0:00:36.529,0:00:38.601
to search for antiviral compounds

0:00:38.625,0:00:41.303
that could help us fight[br]the current pandemic.

0:00:42.080,0:00:45.982
Machine learning is turning[br]the traditional model of drug discovery

0:00:46.006,0:00:47.410
on its head.

0:00:47.434,0:00:48.659
With this approach,

0:00:48.683,0:00:52.761
instead of painstakingly testing[br]thousands of existing molecules

0:00:52.785,0:00:54.221
one by one in a lab

0:00:54.245,0:00:55.832
for their effectiveness,

0:00:55.856,0:01:00.513
we can train a computer[br]to explore the exponentially larger space

0:01:00.537,0:01:04.121
of essentially all possible molecules[br]that could be synthesized,

0:01:04.145,0:01:09.759
and thus, instead of looking[br]for a needle in a haystack,

0:01:09.783,0:01:13.543
we can use the giant magnet[br]of computing power

0:01:13.567,0:01:17.482
to find many needles[br]in multiple haystacks simultaneously.

0:01:18.423,0:01:20.415
We've already had some early success.

0:01:21.010,0:01:26.475
Recently, we used machine learning[br]to discover new antibiotics

0:01:26.499,0:01:29.059
that can help us fight off[br]the bacterial infections

0:01:29.083,0:01:32.694
that can occur alongside[br]SARS-CoV-2 infections.

0:01:33.181,0:01:37.350
Two months ago, TED's Audacious Project[br]approved funding for us

0:01:37.374,0:01:39.562
to massively scale up our work

0:01:39.586,0:01:44.214
with the goal of discovering[br]seven new classes of antibiotics

0:01:44.238,0:01:47.721
against seven of the world's[br]deadly bacterial pathogens

0:01:47.745,0:01:49.800
over the next seven years.

0:01:50.206,0:01:51.939
For context:

0:01:51.963,0:01:53.891
the number of new class of antibiotics

0:01:53.915,0:01:57.150
that have been discovered[br]over the last three decades is zero.

0:01:58.030,0:02:01.601
While the quest for new antibiotics[br]is for our medium-term future,

0:02:01.625,0:02:06.277
the novel coronavirus poses[br]an immediate deadly threat,

0:02:06.301,0:02:10.094
and I'm excited to share that we think[br]we can use the same technology

0:02:10.118,0:02:12.927
to search for therapeutics[br]to fight this virus.

0:02:13.486,0:02:15.205
So how are we going to do it?

0:02:15.229,0:02:18.177
Well, we're creating[br]a compound training library

0:02:18.201,0:02:23.743
and with collaborators applying these[br]molecules to SARS-CoV-2-infected cells

0:02:23.767,0:02:27.661
to see which of them exhibit[br]effective activity.

0:02:28.175,0:02:31.367
These data will be use to train[br]a machine learning model

0:02:31.391,0:02:35.461
that will be applied to an in silico[br]library of over a billion molecules

0:02:35.485,0:02:39.689
to search for potential[br]novel antiviral compounds.

0:02:40.324,0:02:42.982
We will synthesize and test[br]the top predictions

0:02:43.006,0:02:45.895
and advance the most promising[br]candidates into the clinic.

0:02:46.356,0:02:48.134
Sound too good to be true?

0:02:48.158,0:02:49.590
Well, it shouldn't.

0:02:49.614,0:02:52.939
The Antibiotics AI Project is founded[br]on our proof of concept research

0:02:52.963,0:02:56.364
that led to the discovery[br]of a novel broad-spectrum antibiotic

0:02:56.388,0:02:57.573
called halicin.

0:02:58.443,0:03:01.256
Halicin has potent antibacterial activity

0:03:01.280,0:03:05.382
against almost all antibiotic-resistant[br]bacterial pathogens,

0:03:05.406,0:03:09.047
including untreatable[br]panresistant infections.

0:03:09.862,0:03:12.132
Importantly, in contrast[br]to current antibiotics,

0:03:12.156,0:03:15.850
the frequency at which bacteria[br]develop resistance against halicin

0:03:15.874,0:03:17.358
is remarkably low.

0:03:18.303,0:03:23.013
We tested the ability of bacteria[br]to evolve resistance against halicin

0:03:23.037,0:03:24.825
as well as Cipro in the lab.

0:03:25.299,0:03:26.841
In the case of Cipro,

0:03:26.865,0:03:29.690
after just one day, we saw resistance.

0:03:30.213,0:03:31.691
In the case of halicin,

0:03:31.715,0:03:33.830
after one day,[br]we didn't see any resistance.

0:03:34.479,0:03:37.781
Amazingly, after even 30 days,

0:03:37.805,0:03:40.406
we didn't see any resistance[br]against halicin.

0:03:41.098,0:03:46.624
In this pilot project, we first tested[br]roughly 2,500 compounds against E. coli.

0:03:47.259,0:03:50.039
This training set included[br]known antibiotics,

0:03:50.063,0:03:51.809
such as Cipro and penicillin,

0:03:51.833,0:03:54.105
as well as many drugs[br]that are not antibiotics.

0:03:54.984,0:03:57.571
These data we used to train a model

0:03:57.595,0:04:01.573
to learn molecular features[br]associated with antibacterial activity.

0:04:02.269,0:04:04.970
We then applied this model[br]to a drug-repurposing library

0:04:04.994,0:04:07.472
consisting of several thousand molecules

0:04:07.496,0:04:10.114
and asked the model to identify molecules

0:04:10.138,0:04:12.922
that are predicted[br]to have antibacterial properties

0:04:12.946,0:04:15.419
but don't look like existing antibiotics.

0:04:16.427,0:04:21.224
Interestingly, only one molecule[br]in that library fit these criteria,

0:04:21.248,0:04:23.584
and that molecule[br]turned out to be halicin.

0:04:24.444,0:04:27.532
Given that halicin does not look[br]like any existing antibiotic,

0:04:27.556,0:04:31.710
it would have been impossible for a human,[br]including an antibiotic expert,

0:04:31.734,0:04:33.918
to identify halicin in this manner.

0:04:34.574,0:04:37.204
Imagine now what we could do[br]with this technology

0:04:37.228,0:04:38.969
against SARS-CoV-2.

0:04:39.783,0:04:41.148
And that's not all.

0:04:41.172,0:04:43.992
We're also using the tools[br]of synthetic biology,

0:04:44.016,0:04:46.627
tinkering with DNA[br]and other cellular machinery,

0:04:46.651,0:04:50.561
to serve human purposes[br]like combating COVID-19,

0:04:50.585,0:04:54.232
and of note, we are working[br]to develop a protective mask

0:04:54.256,0:04:57.688
that can also serve[br]as a rapid diagnostic test.

0:04:58.192,0:04:59.664
So how does that work?

0:04:59.688,0:05:00.893
Well, we recently showed

0:05:00.917,0:05:03.860
that you can take the cellular[br]machinery out of a living cell

0:05:03.884,0:05:07.976
and freeze-dry it along with[br]RNA sensors onto paper

0:05:08.000,0:05:12.916
in order to create low-cost[br]diagnostics for Ebola and Zika.

0:05:13.503,0:05:18.730
The sensors are activated when[br]they're rehydrated by a patient sample

0:05:18.754,0:05:21.576
that could consist of blood[br]or saliva, for example.

0:05:21.600,0:05:24.861
It turns out, this technology[br]is not limited to paper

0:05:24.885,0:05:27.771
and can be applied[br]to other materials, including cloth.

0:05:28.671,0:05:30.613
For the COVID-19 pandemic,

0:05:30.637,0:05:34.983
we're designing RNA sensors[br]to detect the virus

0:05:35.007,0:05:38.217
and freeze-drying these[br]along with the needed cellular machinery

0:05:38.241,0:05:40.948
into the fabric of a face mask,

0:05:40.972,0:05:43.201
where the simple act of breathing,

0:05:43.225,0:05:45.502
along with the water vapor[br]that comes with it,

0:05:45.526,0:05:47.286
can activate the test.

0:05:47.804,0:05:52.064
Thus, if a patient is infected[br]with SARS-CoV-2,

0:05:52.088,0:05:54.161
the mask will produce[br]a fluorescent signal

0:05:54.185,0:05:58.015
that could be detected by a simple,[br]inexpensive handheld device.

0:05:58.534,0:06:03.018
In one or two hours, a patient[br]could thus be diagnosed

0:06:03.042,0:06:06.014
safely, remotely and accurately.

0:06:06.735,0:06:09.255
We're also using synthetic biology

0:06:09.279,0:06:11.999
to design a candidate[br]vaccine for COVID-19.

0:06:13.014,0:06:15.667
We are repurposing the BCG vaccine,

0:06:15.691,0:06:18.561
which had been used against TB[br]for almost a century.

0:06:18.585,0:06:20.126
It's a live attenuated vaccine,

0:06:20.150,0:06:24.807
and we're engineering it[br]to express SARS-CoV-2 antigens,

0:06:24.831,0:06:27.645
which should trigger the production[br]of protective antibodies

0:06:27.669,0:06:29.304
by the immune system.

0:06:29.328,0:06:32.062
Importantly, BCG[br]is massively scalable

0:06:32.086,0:06:36.659
and has a safety profile that's among[br]the best of any reported vaccine.

0:06:37.881,0:06:42.986
With the tools of synthetic biology[br]and artificial intelligence,

0:06:43.010,0:06:46.358
we can win the fight[br]against this novel coronavirus.

0:06:46.844,0:06:50.163
This work is in its very early stages,[br]but the promise is real.

0:06:50.798,0:06:54.243
Science and technology[br]can give us an important advantage

0:06:54.267,0:06:57.428
in the battle of human wits[br]versus the genes of superbugs,

0:06:57.452,0:06:59.199
a battle we can win.

0:06:59.990,0:07:01.223
Thank you.