0:00:00.750,0:00:05.259
The second half of the last century[br]was completely defined

0:00:05.283,0:00:07.282
by a technological revolution:

0:00:07.306,0:00:08.741
the software revolution.

0:00:09.313,0:00:14.121
The ability to program electrons[br]on a material called silicon

0:00:14.145,0:00:17.218
made possible technologies,[br]companies and industries

0:00:17.242,0:00:21.219
that were at one point[br]unimaginable to many of us,

0:00:21.243,0:00:25.158
but which have now fundamentally changed[br]the way the world works.

0:00:26.158,0:00:28.079
The first half of this century, though,

0:00:28.103,0:00:32.081
is going to be transformed[br]by a new software revolution:

0:00:32.105,0:00:34.540
the living software revolution.

0:00:34.921,0:00:38.971
And this will be powered by the ability[br]to program biochemistry

0:00:38.995,0:00:41.290
on a material called biology.

0:00:41.314,0:00:45.455
And doing so will enable us to harness[br]the properties of biology

0:00:45.479,0:00:48.135
to generate new kinds of therapies,

0:00:48.159,0:00:50.027
to repair damaged tissue,

0:00:50.051,0:00:52.776
to reprogram faulty cells

0:00:52.800,0:00:57.354
or even build programmable[br]operating systems out of biochemistry.

0:00:58.420,0:01:01.993
If we can realize this --[br]and we do need to realize it --

0:01:02.017,0:01:04.179
its impact will be so enormous

0:01:04.203,0:01:08.080
that it will make the first[br]software revolution pale in comparison.

0:01:08.104,0:01:12.338
And that's because living software[br]would transform the entirety of medicine,

0:01:12.362,0:01:13.921
agriculture and energy,

0:01:13.945,0:01:17.773
and these are sectors that dwarf[br]those dominated by IT.

0:01:18.812,0:01:22.986
Imagine programmable plants[br]that fix nitrogen more effectively

0:01:23.010,0:01:25.915
or resist emerging fungal pathogens,

0:01:25.939,0:01:29.476
or even programming crops[br]to be perennial rather than annual

0:01:29.500,0:01:31.768
so you could double[br]your crop yields each year.

0:01:31.792,0:01:33.890
That would transform agriculture

0:01:33.914,0:01:38.018
and how we'll keep our growing[br]and global population fed.

0:01:38.794,0:01:41.056
Or imagine programmable immunity,

0:01:41.080,0:01:45.318
designing and harnessing molecular devices[br]that guide your immune system

0:01:45.342,0:01:49.172
to detect, eradicate[br]or even prevent disease.

0:01:49.196,0:01:50.767
This would transform medicine

0:01:50.791,0:01:54.280
and how we'll keep our growing[br]and aging population healthy.

0:01:55.501,0:01:59.704
We already have many of the tools[br]that will make living software a reality.

0:01:59.728,0:02:02.075
We can precisely edit genes with CRISPR.

0:02:02.099,0:02:05.182
We can rewrite the genetic code[br]one base at a time.

0:02:05.206,0:02:09.642
We can even build functioning[br]synthetic circuits out of DNA.

0:02:10.428,0:02:12.897
But figuring out how and when[br]to wield these tools

0:02:12.921,0:02:15.343
is still a process of trial and error.

0:02:15.367,0:02:19.027
It needs deep expertise,[br]years of specialization.

0:02:19.051,0:02:22.088
And experimental protocols[br]are difficult to discover

0:02:22.112,0:02:24.694
and all too often, difficult to reproduce.

0:02:25.256,0:02:29.729
And, you know, we have a tendency[br]in biology to focus a lot on the parts,

0:02:29.753,0:02:32.886
but we all know that something like flying[br]wouldn't be understood

0:02:32.910,0:02:34.249
by only studying feathers.

0:02:34.846,0:02:39.367
So programming biology is not yet[br]as simple as programming your computer.

0:02:39.391,0:02:41.069
And then to make matters worse,

0:02:41.093,0:02:45.103
living systems largely bear no resemblance[br]to the engineered systems

0:02:45.127,0:02:47.223
that you and I program every day.

0:02:47.691,0:02:51.802
In contrast to engineered systems,[br]living systems self-generate,

0:02:51.826,0:02:53.297
they self-organize,

0:02:53.321,0:02:55.008
they operate at molecular scales.

0:02:55.032,0:02:57.168
And these molecular-level interactions

0:02:57.192,0:03:00.210
lead generally to robust[br]macro-scale output.

0:03:00.234,0:03:02.954
They can even self-repair.

0:03:04.256,0:03:07.250
Consider, for example,[br]the humble household plant,

0:03:07.274,0:03:09.461
like that one sat[br]on your mantelpiece at home

0:03:09.485,0:03:11.272
that you keep forgetting to water.

0:03:11.749,0:03:15.364
Every day, despite your neglect,[br]that plant has to wake up

0:03:15.388,0:03:18.135
and figure out how[br]to allocate its resources.

0:03:18.159,0:03:21.730
Will it grow, photosynthesize,[br]produce seeds, or flower?

0:03:21.754,0:03:25.693
And that's a decision that has to be made[br]at the level of the whole organism.

0:03:25.717,0:03:29.198
But a plant doesn't have a brain[br]to figure all of that out.

0:03:29.222,0:03:31.939
It has to make do[br]with the cells on its leaves.

0:03:31.963,0:03:33.866
They have to respond to the environment

0:03:33.890,0:03:36.539
and make the decisions[br]that affect the whole plant.

0:03:36.563,0:03:40.551
So somehow there must be a program[br]running inside these cells,

0:03:40.575,0:03:43.302
a program that responds[br]to input signals and cues

0:03:43.326,0:03:45.266
and shapes what that cell will do.

0:03:45.679,0:03:48.926
And then those programs must operate[br]in a distributed way

0:03:48.950,0:03:50.287
across individual cells,

0:03:50.311,0:03:54.434
so that they can coordinate[br]and that plant can grow and flourish.

0:03:55.675,0:03:58.991
If we could understand[br]these biological programs,

0:03:59.015,0:04:02.137
if we could understand[br]biological computation,

0:04:02.161,0:04:06.098
it would transform our ability[br]to understand how and why

0:04:06.122,0:04:07.668
cells do what they do.

0:04:08.152,0:04:10.139
Because, if we understood these programs,

0:04:10.163,0:04:12.296
we could debug them when things go wrong.

0:04:12.320,0:04:16.513
Or we could learn from them how to design[br]the kind of synthetic circuits

0:04:16.537,0:04:21.011
that truly exploit[br]the computational power of biochemistry.

0:04:22.407,0:04:25.425
My passion about this idea[br]led me to a career in research

0:04:25.449,0:04:29.080
at the interface of maths,[br]computer science and biology.

0:04:29.104,0:04:33.830
And in my work, I focus on the concept[br]of biology as computation.

0:04:34.334,0:04:37.476
And that means asking[br]what do cells compute,

0:04:37.500,0:04:41.017
and how can we uncover[br]these biological programs?

0:04:41.760,0:04:45.517
And I started to ask these questions[br]together with some brilliant collaborators

0:04:45.541,0:04:48.112
at Microsoft Research[br]and the University of Cambridge,

0:04:48.136,0:04:50.419
where together we wanted to understand

0:04:50.443,0:04:54.620
the biological program[br]running inside a unique type of cell:

0:04:54.644,0:04:56.538
an embryonic stem cell.

0:04:57.136,0:05:00.296
These cells are unique[br]because they're totally naïve.

0:05:00.320,0:05:02.488
They can become anything they want:

0:05:02.512,0:05:05.077
a brain cell, a heart cell,[br]a bone cell, a lung cell,

0:05:05.101,0:05:06.998
any adult cell type.

0:05:07.022,0:05:08.699
This naïvety, it sets them apart,

0:05:08.723,0:05:11.724
but it also ignited the imagination[br]of the scientific community,

0:05:11.748,0:05:15.011
who realized, if we could[br]tap into that potential,

0:05:15.035,0:05:17.386
we would have a powerful[br]tool for medicine.

0:05:17.917,0:05:20.538
If we could figure out[br]how these cells make the decision

0:05:20.562,0:05:22.693
to become one cell type or another,

0:05:22.717,0:05:24.407
we might be able to harness them

0:05:24.431,0:05:28.984
to generate cells that we need[br]to repair diseased or damaged tissue.

0:05:29.794,0:05:32.724
But realizing that vision[br]is not without its challenges,

0:05:32.748,0:05:35.512
not least because these particular cells,

0:05:35.536,0:05:38.365
they emerge just six days[br]after conception.

0:05:38.826,0:05:40.881
And then within a day or so, they're gone.

0:05:40.905,0:05:42.962
They have set off down the different paths

0:05:42.986,0:05:46.036
that form all the structures[br]and organs of your adult body.

0:05:47.770,0:05:50.849
But it turns out that cell fates[br]are a lot more plastic

0:05:50.873,0:05:52.286
than we might have imagined.

0:05:52.310,0:05:56.631
About 13 years ago, some scientists[br]showed something truly revolutionary.

0:05:57.393,0:06:01.739
By inserting just a handful of genes[br]into an adult cell,

0:06:01.763,0:06:03.527
like one of your skin cells,

0:06:03.551,0:06:07.510
you can transform that cell[br]back to the naïve state.

0:06:07.534,0:06:10.709
And it's a process that's actually[br]known as "reprogramming,"

0:06:10.733,0:06:14.092
and it allows us to imagine[br]a kind of stem cell utopia,

0:06:14.116,0:06:17.757
the ability to take a sample[br]of a patient's own cells,

0:06:17.781,0:06:20.141
transform them back to the naïve state

0:06:20.165,0:06:23.295
and use those cells to make[br]whatever that patient might need,

0:06:23.319,0:06:25.394
whether it's brain cells or heart cells.

0:06:26.541,0:06:28.306
But over the last decade or so,

0:06:28.330,0:06:31.374
figuring out how to change cell fate,

0:06:31.398,0:06:33.550
it's still a process of trial and error.

0:06:33.911,0:06:38.419
Even in cases where we've uncovered[br]successful experimental protocols,

0:06:38.443,0:06:39.910
they're still inefficient,

0:06:39.934,0:06:44.172
and we lack a fundamental understanding[br]of how and why they work.

0:06:44.650,0:06:47.655
If you figured out how to change[br]a stem cell into a heart cell,

0:06:47.679,0:06:50.768
that hasn't got any way of telling you[br]how to change a stem cell

0:06:50.792,0:06:51.993
into a brain cell.

0:06:52.633,0:06:55.564
So we wanted to understand[br]the biological program

0:06:55.588,0:06:58.035
running inside an embryonic stem cell,

0:06:58.059,0:07:01.565
and understanding the computation[br]performed by a living system

0:07:01.589,0:07:05.842
starts with asking[br]a devastatingly simple question:

0:07:05.866,0:07:09.222
What is it that system actually has to do?

0:07:09.838,0:07:12.688
Now, computer science actually[br]has a set of strategies

0:07:12.712,0:07:16.539
for dealing with what it is the software[br]and hardware are meant to do.

0:07:16.563,0:07:19.223
When you write a program,[br]you code a piece of software,

0:07:19.247,0:07:21.247
you want that software to run correctly.

0:07:21.271,0:07:23.061
You want performance, functionality.

0:07:23.085,0:07:24.302
You want to prevent bugs.

0:07:24.326,0:07:25.634
They can cost you a lot.

0:07:26.168,0:07:28.010
So when a developer writes a program,

0:07:28.034,0:07:30.304
they could write down[br]a set of specifications.

0:07:30.328,0:07:32.199
These are what your program should do.

0:07:32.223,0:07:34.491
Maybe it should compare[br]the size of two numbers

0:07:34.515,0:07:36.307
or order numbers by increasing size.

0:07:37.037,0:07:41.732
Technology exists that allows us[br]automatically to check

0:07:41.756,0:07:44.134
whether our specifications are satisfied,

0:07:44.158,0:07:46.791
whether that program[br]does what it should do.

0:07:47.266,0:07:50.122
And so our idea was that in the same way,

0:07:50.146,0:07:53.214
experimental observations,[br]things we measure in the lab,

0:07:53.238,0:07:58.271
they correspond to specifications[br]of what the biological program should do.

0:07:58.769,0:08:00.645
So we just needed to figure out a way

0:08:00.669,0:08:03.852
to encode this new type of specification.

0:08:04.594,0:08:08.248
So let's say you've been busy in the lab[br]and you've been measuring your genes

0:08:08.272,0:08:10.708
and you've found that if Gene A is active,

0:08:10.732,0:08:14.120
then Gene B or Gene C seems to be active.

0:08:14.678,0:08:18.260
We can write that observation down[br]as a mathematical expression

0:08:18.284,0:08:20.657
if we can use the language of logic:

0:08:21.125,0:08:23.453
If A, then B or C.

0:08:24.242,0:08:26.696
Now, this is a very simple example, OK.

0:08:26.720,0:08:28.463
It's just to illustrate the point.

0:08:28.487,0:08:31.411
We can encode truly rich expressions

0:08:31.435,0:08:35.588
that actually capture the behavior[br]of multiple genes or proteins over time

0:08:35.612,0:08:38.148
across multiple different experiments.

0:08:38.521,0:08:41.147
And so by translating our observations

0:08:41.171,0:08:43.164
into mathematical expression in this way,

0:08:43.188,0:08:48.286
it becomes possible to test whether[br]or not those observations can emerge

0:08:48.310,0:08:51.364
from a program of genetic interactions.

0:08:52.063,0:08:54.619
And we developed a tool to do just this.

0:08:54.643,0:08:57.525
We were able to use this tool[br]to encode observations

0:08:57.549,0:08:58.956
as mathematical expressions,

0:08:58.980,0:09:02.590
and then that tool would allow us[br]to uncover the genetic program

0:09:02.614,0:09:04.152
that could explain them all.

0:09:05.481,0:09:07.761
And we then apply this approach

0:09:07.785,0:09:11.868
to uncover the genetic program[br]running inside embryonic stem cells

0:09:11.892,0:09:16.081
to see if we could understand[br]how to induce that naïve state.

0:09:16.105,0:09:18.057
And this tool was actually built

0:09:18.081,0:09:20.733
on a solver that's deployed[br]routinely around the world

0:09:20.757,0:09:23.026
for conventional software verification.

0:09:23.630,0:09:27.321
So we started with a set[br]of nearly 50 different specifications

0:09:27.345,0:09:31.851
that we generated from experimental[br]observations of embryonic stem cells.

0:09:31.875,0:09:34.511
And by encoding these[br]observations in this tool,

0:09:34.535,0:09:37.720
we were able to uncover[br]the first molecular program

0:09:37.744,0:09:39.705
that could explain all of them.

0:09:40.309,0:09:42.822
Now, that's kind of a feat[br]in and of itself, right?

0:09:42.846,0:09:45.748
Being able to reconcile[br]all of these different observations

0:09:45.772,0:09:48.839
is not the kind of thing[br]you can do on the back of an envelope,

0:09:48.863,0:09:51.511
even if you have a really big envelope.

0:09:52.190,0:09:54.348
Because we've got[br]this kind of understanding,

0:09:54.372,0:09:55.834
we could go one step further.

0:09:55.858,0:09:59.229
We could use this program to predict[br]what this cell might do

0:09:59.253,0:10:01.429
in conditions we hadn't yet tested.

0:10:01.453,0:10:03.854
We could probe the program in silico.

0:10:04.735,0:10:05.982
And so we did just that:

0:10:06.006,0:10:09.186
we generated predictions[br]that we tested in the lab,

0:10:09.210,0:10:12.242
and we found that this program[br]was highly predictive.

0:10:12.266,0:10:14.891
It told us how we could[br]accelerate progress

0:10:14.915,0:10:17.975
back to the naïve state[br]quickly and efficiently.

0:10:17.999,0:10:20.569
It told us which genes[br]to target to do that,

0:10:20.593,0:10:23.217
which genes might even[br]hinder that process.

0:10:23.241,0:10:28.231
We even found the program predicted[br]the order in which genes would switch on.

0:10:28.980,0:10:32.120
So this approach really allowed us[br]to uncover the dynamics

0:10:32.144,0:10:34.546
of what the cells are doing.

0:10:35.728,0:10:39.370
What we've developed, it's not a method[br]that's specific to stem cell biology.

0:10:39.394,0:10:42.078
Rather, it allows us to make sense[br]of the computation

0:10:42.102,0:10:43.787
being carried out by the cell

0:10:43.811,0:10:46.642
in the context of genetic interactions.

0:10:46.666,0:10:48.954
So really, it's just one building block.

0:10:48.978,0:10:51.663
The field urgently needs[br]to develop new approaches

0:10:51.687,0:10:54.382
to understand biological[br]computation more broadly

0:10:54.406,0:10:55.773
and at different levels,

0:10:55.797,0:10:59.926
from DNA right through[br]to the flow of information between cells.

0:10:59.950,0:11:02.747
Only this kind of[br]transformative understanding

0:11:02.771,0:11:07.757
will enable us to harness biology[br]in ways that are predictable and reliable.

0:11:09.029,0:11:12.071
But to program biology,[br]we will also need to develop

0:11:12.095,0:11:14.090
the kinds of tools and languages

0:11:14.114,0:11:17.522
that allow both experimentalists[br]and computational scientists

0:11:17.546,0:11:20.043
to design biological function

0:11:20.067,0:11:23.572
and have those designs compile down[br]to the machine code of the cell,

0:11:23.596,0:11:24.777
its biochemistry,

0:11:24.801,0:11:27.285
so that we could then[br]build those structures.

0:11:27.309,0:11:30.982
Now, that's something akin[br]to a living software compiler,

0:11:31.006,0:11:33.222
and I'm proud to be[br]part of a team at Microsoft

0:11:33.246,0:11:34.898
that's working to develop one.

0:11:35.366,0:11:38.592
Though to say it's a grand challenge[br]is kind of an understatement,

0:11:38.616,0:11:39.789
but if it's realized,

0:11:39.813,0:11:43.522
it would be the final bridge[br]between software and wetware.

0:11:45.006,0:11:48.421
More broadly, though, programming biology[br]is only going to be possible

0:11:48.445,0:11:52.724
if we can transform the field[br]into being truly interdisciplinary.

0:11:52.748,0:11:55.700
It needs us to bridge[br]the physical and the life sciences,

0:11:55.724,0:11:57.991
and scientists from[br]each of these disciplines

0:11:58.015,0:12:00.746
need to be able to work together[br]with common languages

0:12:00.770,0:12:03.489
and to have shared scientific questions.

0:12:04.757,0:12:08.750
In the long term, it's worth remembering[br]that many of the giant software companies

0:12:08.774,0:12:11.266
and the technology[br]that you and I work with every day

0:12:11.290,0:12:12.793
could hardly have been imagined

0:12:12.817,0:12:16.422
at the time we first started[br]programming on silicon microchips.

0:12:16.446,0:12:19.477
And if we start now to think about[br]the potential for technology

0:12:19.501,0:12:21.927
enabled by computational biology,

0:12:21.951,0:12:24.886
we'll see some of the steps[br]that we need to take along the way

0:12:24.910,0:12:26.343
to make that a reality.

0:12:27.231,0:12:30.313
Now, there is the sobering thought[br]that this kind of technology

0:12:30.337,0:12:32.114
could be open to misuse.

0:12:32.138,0:12:34.301
If we're willing to talk[br]about the potential

0:12:34.325,0:12:35.761
for programming immune cells,

0:12:35.785,0:12:38.973
we should also be thinking[br]about the potential of bacteria

0:12:38.997,0:12:40.658
engineered to evade them.

0:12:40.682,0:12:42.769
There might be people willing to do that.

0:12:43.506,0:12:45.228
Now, one reassuring thought in this

0:12:45.252,0:12:47.541
is that -- well, less so[br]for the scientists --

0:12:47.565,0:12:50.834
is that biology is[br]a fragile thing to work with.

0:12:50.858,0:12:53.270
So programming biology[br]is not going to be something

0:12:53.294,0:12:55.142
you'll be doing in your garden shed.

0:12:55.642,0:12:57.722
But because we're at the outset of this,

0:12:57.746,0:13:00.329
we can move forward[br]with our eyes wide open.

0:13:00.353,0:13:02.677
We can ask the difficult[br]questions up front,

0:13:02.701,0:13:05.741
we can put in place[br]the necessary safeguards

0:13:05.765,0:13:08.562
and, as part of that,[br]we'll have to think about our ethics.

0:13:08.586,0:13:11.758
We'll have to think about putting bounds[br]on the implementation

0:13:11.782,0:13:13.280
of biological function.

0:13:13.604,0:13:17.319
So as part of this, research in bioethics[br]will have to be a priority.

0:13:17.343,0:13:19.750
It can't be relegated to second place

0:13:19.774,0:13:22.288
in the excitement[br]of scientific innovation.

0:13:23.154,0:13:26.628
But the ultimate prize,[br]the ultimate destination on this journey,

0:13:26.652,0:13:30.096
would be breakthrough applications[br]and breakthrough industries

0:13:30.120,0:13:33.564
in areas from agriculture and medicine[br]to energy and materials

0:13:33.588,0:13:35.849
and even computing itself.

0:13:36.490,0:13:39.638
Imagine, one day we could be powering[br]the planet sustainably

0:13:39.662,0:13:41.521
on the ultimate green energy

0:13:41.545,0:13:45.488
if we could mimic something[br]that plants figured out millennia ago:

0:13:45.512,0:13:49.283
how to harness the sun's energy[br]with an efficiency that is unparalleled

0:13:49.307,0:13:51.163
by our current solar cells.

0:13:51.695,0:13:54.296
If we understood that program[br]of quantum interactions

0:13:54.320,0:13:57.584
that allow plants to absorb[br]sunlight so efficiently,

0:13:57.608,0:14:01.552
we might be able to translate that[br]into building synthetic DNA circuits

0:14:01.576,0:14:04.489
that offer the material[br]for better solar cells.

0:14:05.349,0:14:09.042
There are teams and scientists working[br]on the fundamentals of this right now,

0:14:09.066,0:14:12.309
so perhaps if it got the right attention[br]and the right investment,

0:14:12.333,0:14:14.613
it could be realized in 10 or 15 years.

0:14:15.457,0:14:18.654
So we are at the beginning[br]of a technological revolution.

0:14:19.067,0:14:22.288
Understanding this ancient type[br]of biological computation

0:14:22.312,0:14:24.444
is the critical first step.

0:14:24.468,0:14:25.785
And if we can realize this,

0:14:25.809,0:14:28.651
we would enter in the era[br]of an operating system

0:14:28.675,0:14:30.580
that runs living software.

0:14:30.604,0:14:31.770
Thank you very much.

0:14:31.794,0:14:34.484
(Applause)