0:58
Hi everybody.
0:59
Will, the CEO and Co-founder of Pando Bio, we're located in Watertown, mainly using generative AI to discover enzyme and engineer enzymes for pharma manufacturing and diagnostics.
1:12
So we make a lot of enzymes for sequencing application.
1:17
This is our platform basically using AI to do high load mutation to optimise multiple properties simultaneously and then we synthesise this DNA for the enzyme in the third vendor receives it and we have an in house ultra-high throughput screening platform that can screen 1000 times more enzymes compared to our competitors.
1:41
And then these high quality and quantity data will be returned back to the AI for continuous learning.
1:48
So we have engineered a whole platform to be only 24 days, which is much faster compared to your traditional enzyme engineering company.
1:58
So as you may see, we are not only doing in silico but also experimental validation.
2:04
Then you may be asking how are you going to screen 1,000 genotypes at a time.
2:09
So this is our in house ultra-high throughput screening platform.
2:13
It is droplet based.
2:14
So what's happened is there are billions of droplets in one sample and there will be 0 to 1 cell harbouring a unique AI design enzyme in each droplet.
2:27
And then when we apply different selective pressure on each droplet, the enzyme would release.
2:35
And then if the enzyme is behaving very well, it will replicate its own DNA.
2:41
If it is a less active enzyme, it will expand its DNA in the lesser extent. Later on, when we merge the droplets, we just need to do a sequencing round.
2:51
We can get an enrichment of how well the enzyme is behaving right.
2:57
So using these data from the high throughput screening, we train our AI model with these in house data.
3:04
So you can see here it's the functional heat and you can see in just two rounds, the AI will give you roughly tenfold of increase on the heat.
3:15
And these like these multiple properties are all engineered in one round by the AI.
3:26
And if we're just using activity as a property to look at, you can see that compared to the benchmark, AI can do a very good job, 2.5 fold compared to the benchmark.
3:42
And it's 22 times better than the non AI method, just like random mutagenesis and in the second round, and it can even increase to five times more than the benchmark.
3:53
So AI provides a really effective tool.
3:55
But again, like I said, if you don't have an experiment validation, it's very hard to get a very good increase in multiple rounds.
4:05
So we've been on the market for roughly 2 years, a very young startup of five people, but we already demonstrated our customer proposition in a real customer collaboration case.
4:19
For example, in the first one, we are working with a pharma manufacturing customer, they are making an active pharmaceutical ingredient, 11 chemical steps was replaced by 1 enzymatic step and we discovered this new enzyme for them that basically help them drop the manufacturing steps, therefore increasing $24 million revenue per year.
4:43
In another case, we help with a diagnostic customer with the enzyme basically to make them more thermal stable and also sensitive.
4:52
So you can see they got a lot of our new revenue increase due to the serviceable market expansion.
5:00
All right, so the kind of collaboration or services we offer are basically twofold.
5:07
One is Enzyme discovery.
5:09
So imagine if you were in this mountain range, right?
5:13
You want to find the trail heading to the tallest peak.
5:17
So enzyme discovery is basically helping you.
5:20
Where's my trail hat?
5:22
And then enzyme engineering is like, OK, I got my trail hat, how do I even climb to that mountain?
5:28
So enzyme engineering is helping you to improve the variant to climb to that summit.
5:34
And here I'm going to demonstrate multiple cases.
5:38
So this one is a case working with a diagnostic customer of finding a novel family B polymerase for NGS lab prep.
5:48
And so everybody know hi-fi polymerase well studied very crowded IP.
5:55
So that's the first thing is we want to make sure that we're finding some enzyme that is far away that is not infringing any existing IP.
6:03
And of course we want it to be high thermal stable, processivity is good, speed is good, fidelity is good.
6:11
So thanks to our very unique database, which is five times larger than a UniProt.
6:18
So it consists of our internal generated data.
6:22
At the same time, well curated public data, for example, include a lot of metagenomes from these extreme environments like hot spring, volcano and deep sea habitats and there's no fragments or short sequences.
6:37
So these databases serve as the rich resources for our AI model.
6:43
And you can see in just one round of engineering, all these dots are showing our AI engineered enzyme and being tested, we measured activity, and you can see these two blue are PFU and KOD basically.
6:57
And the green dots are the enzymes that have very high activity and the orange dots are not only high activity but also high thermal stability.
7:08
So because we want to invent new IP.
7:11
So our goal is basically identifying new orange dots that lay far apart from this market leader cluster.
7:19
So your sequence is very different.
7:22
And you can see here that is validated by our collaborator on the PCR validation that our 4% of the wrong one AI design enzyme already are marching the market leader and they are all FTO and we have already completed this project by in two rounds.
7:41
I'm not showing more data due to some contract reason.
7:46
The second case is basically engineering a heavily patented enzyme AKA reverse transcriptase with another collaborator.
7:57
And again, the goal is basically to have high thermal stability, processivity speed and of course RT activity to match.
8:07
You know, the market leader kind of like everybody probably uses it from Thermo Fisher, not going to name it, but anyway, still want to be IP-free.
8:15
And if you work with RT, you'll probably know super heavily patented enzyme, almost 60% of the positions already being patented.
8:24
So how are you going to find novel enzyme in this 40?
8:28
Probably going to be impossible if you're doing random mutagenesis, right?
8:33
So what we do is we use AI to learn from these patterns and literature, but the AI will make a mutation to avoid infringing those IPs.
8:45
And you can see here, this is just one round.
8:48
We already finding a hit that is matching the market leader at 60 Celsius, which is the working condition for the best enzyme currently on the market.
9:02
So those are the ones that kind of like the enzyme solutions we work with collaborator.
9:07
However, I'm going to show you some internal assets that we're developing.
9:11
We own the IP, it's called Phi29.
9:16
So again used a lot in RCA and whole genome sequencing.
9:19
If you're here, a lot of people use Phi29.
9:22
We have an engineered enzymes to make 100% call a plasmid assembly from 100% of the sequenced colony and it is 6 times faster in the reaction speed.
9:35
At the same time extremely low GC bias.
9:38
We have a tested on the blend of genome, a bacteria genome or human genome and you can see our enzymes have very low GC bias.
9:47
At the same time, we engineered an enzyme to make basically the A soluble protein compared to the total protein from zero, almost zero to 100%.
10:01
And by the way, we have engineered enzyme to be so good that is transportable at room temperature.
10:07
Actually I have a sample in my batch.
10:10
So if you want some free sample, we have a startup booth right outside near the poster.
10:14
Feel free to grab one in your luggage and then test it.
10:18
And again, located in Watertown, not only Phi29 also are working on ligases and RNA ligase.
10:25
So if you're interested, visit pando.bio that's our website or contact us at the e-mail or talk to us at the booth.
10:34
Thank you.