Generative artificial intelligence empowers digital twins in drug discovery and clinical trials

Maria Bordukova, Nikita Makarov, Raul Rodriguez-Esteban, Fabian Schmich & Michael P. Menden

Steve Cheney

DOI

What is a Digital Twin (DT)?

Generative artificial intelligence empowers digital twins in drug discovery and clinical trials

What is a Digital Twin (DT)?

Generative artificial intelligence empowers digital twins in drug discovery and clinical trials

• As described by Grieves

​

1) A data structure for the real-world system

​

2) Some process that links data together to form dynamics

​

3) Some link to the real world that feeds back data into the data-propagation/generation process

• Expanded by Wright & Davidson

​

1) Sufficiently physics-based that updating parameters within the model based on measurement data is a meaningful thing to do

​

2) Sufficiently accurate that the updated parameter values will be useful for the application of interest

​

3) Sufficiently quick to run that decisions about the application can be made within the required timescale

Fig. 1 - Digital Twins (DTs) have multiple use cases

Generative artificial intelligence empowers digital twins in drug discovery and clinical trials

Fig. 2 - Generative digital twins (DTs) can be realized by various deep learning (DL) architectures.

Generative artificial intelligence empowers digital twins in drug discovery and clinical trials

14

Drugs fully generated by AI have in clinical trials

As of March 2023*

Generative artificial intelligence empowers digital twins in drug discovery and clinical trials

* Reference no longer available online, hard to come by up to date numbers

Fig. 3 - Digital twin (DT) architectures and use cases in clinical trials.

Generative artificial intelligence empowers digital twins in drug discovery and clinical trials

• Flow of info between DTs and patients is bidirectional

​

• Existing generative DT architectures
• (b) conditional restricted Boltzmann machine (CRBM)
• (c) variational autoencoder (VAE)
• Potential generative DT architectures are
• (d) generative adversarial networks (GAN)
• (e) stable diffusion
• (f) neural ordinary differential equations (neural ODE)
• (g) transformers

Digital Twins in preclinical drug discovery

Generative artificial intelligence empowers digital twins in drug discovery and clinical trials

Individual cells

• Commercial offerings in multi-omics-based human cells
• No mention of generative approaches

Cell culture

• Generative DTs developed to create synthetic images and model bacterial growth distribution
• Use of GAN models

Tissue

• Largest development in histopathology
• Used to extract tumor microenvironment (TME) properties
• Generative DTs have been employed to generate synthetic images based on training sets

Organ and organ systems

• Mechanistic models dominate the domain
• Drawbacks: requires prior knowledge of physiological interactions within system

Animal models

• CNN-based DTs proposed to simulate visual response across the mouse visual cortex
• GAN-based rat DTs were developed for hepatotoxicity assessment by predicting the dynamics of clinical parameters and toxicogenomics simulations by generating a liver transcriptomic profile

Digital Twins in clinical drug discovery

Generative artificial intelligence empowers digital twins in drug discovery and clinical trials

• Possibility to reduce trial population
• Motivation for DTs includes the low approval rate of compounds in clinical trials
• Most current DT models use shallow NNs with limited feature learning, unlike deeper NN models which offer more advanced capabilities
• Recent generative AI advances in DTs utilize VAEs for predicting patient trajectories, offering more sophisticated simulation methods
• DT model applicability is limited by focus on single use cases, small training datasets, and basic validation methods

The Gartner hype cycle - Wikimedia

Key takeaways (as suggested by authors)

1. Digital Twins are emerging at all stages of drug discovery and development.
2. Generative artificial intelligence (AI) is poised to be the underlying technology supporting future development of Digital Twins.
3. A number of both academic and commercial Digital Twins can be found in preclinical drug development models.
4. Digital Twins in clinical trials still need further development, both methodological and from a regulatory standpoint.
5. It is anticipated that there are many use cases where Digital Twins will be applied in clinical trials in the near future, including interim trial analysis, adverse event prediction and in trial design.
6. The development of a foundation model may further advance Digital Twins in drug discovery and clinical trials.

xkcd.com/2451

Generative artificial intelligence empowers digital twins in drug discovery and clinical trials