AI > Chemistry and pharma
g
decision-making processes. While the technology has been previously applied in various use cases, it has generally not been applied to biological systems. Raminderpal Singh, founder & CEO at Incubate Bio, explained the use of causal AI, first popularised by Professor Judea Pearl; “What Pearl has done is to take the Bayesian theorem and turned it into a ‘Bayesian network’ and put some logic on top of it. He’s created a nice straightforward path for real causal understanding of systems. This approach has been picked up by a number of universities and organisations including Microsoft, Amazon, IBM research groups that also include social media and FinTech organisations.” “In his book, The Book of Why: The
New Science of Cause and Effect, Professor Pearl explains the notion of understanding and how to go after mechanism modelling such that when you predict what’s happening inside the mechanism, not just as a black box,” said Singh. “Now, the challenge of these things is that many are used to what they call ‘zero prior knowledge’ systems now or ‘knowledge-less’ systems, which is what a black box system is, a black box system assumes nothing.” “People are doing machine learning and AI, they’ve all grown up in traditional statistical thinking, whether they’re working on a simple algorithm, linear regression or some complex deep network. It’s really a model they’re developing based on training data. But then you’ve got all this Bayesian stuff which is untapped,” Singh said. Existing AI companies “will use
Bayesian theorems. But they’ve not thought about this Bayesian network, and how that can allow them to do mechanism modelling,” stressed Singh. “When we started the company this is being done in other industries, we know Pearl’s stuff is being done in other industries. But it’s not being done here. And we realised the reason it’s not to be done previously in biology is because its really hard to do anything in the world of biology really hard.”
Causal AI is challenging to apply to biological systems because there is often a lack of “sufficiency” said Singh. When looking at a potential molecule at what point can a scientists make a statement about whether it can go to a clinical trial or be given to a patient? There’s never enough data in a biological system. So you kind of have to go with a lot of subjectivity. And that creates a lot of nervousness about people’s ability to use AI and make decisions on it, said
20 Scientific Computing World Summer 2023
‘To get to grips with causal AI, biological scientists need to forget machine learning and forget [they] even know any machine learning’
Raminderpal Singh, founder and CEO at Incubate Bio
unique combination of a Large Language Model approach and a powerful proven Causal AI engine. Together these allow ALaSCA to enable early-stage drug discovery groups to rapidly interrogate and extract meaningful biological insights from multiple public research literature and comprehensive datasets, and then perform causal AI on the derived mechanisms. The resulting output provides a unique
Cancer Model System, which can be used by scientists to find directly actionable insights for their drug discovery program. Singh said that as causal AI is very
different from existing approaches, and as such there needs to be some education to get organisations to use these systems. He said: “All these guys that have
Singh: “But we’ve cracked that now in the company. And that’s where we’re pushing hard.” In July, Incubate Bio announced the
commercial release of its ALaSCA software platform. ALaSCA has been developed specifically to enable rapid investigation and in silico experimentation to assist the discovery and development of novel cancer therapies, with an initial focus on the DNA Damage Response (DDR) pathway. As the first commercially available
application of Causal AI technology to the DDR pathway, the latest release of ALaSCA will be especially applicable to developing novel DDR-targeting therapeutics. It will also help support the identification of potential DDR- combination therapies for treating many cancer types, including breast, ovarian and other hard-to-treat, drug-resistant cancers. ALaSCA is underpinned by a
grown up with traditional statistical thinking, you see. They’ve not grown up thinking like this. To get to grips with causal AI, biological scientists need to forget machine learning and forget ‘I even know any machine learning, how do I create a mathematical model’, which is representative of what the scientist believes is the model – where the biological mechanism is the model.” “You really have to understand the
diseases very well and you also need to understand the datasets. Data engineering is even more of a challenge with these datasets. There are missing data. “And you can’t just simply impute data,
you really have to have an understanding when you make a decision around a dirty data set of noisy datasets. It’s very complex. And it’s tied to understanding biology. It’s like drawing a picture with a lack of paints. “With ALaSCA and ALaSCAdb, scientists
can now draw that picture,” Singh concluded. SCW
NESSDesign/
Shutterstock.com
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42
