Page 33 - AIH-1-3

Artificial Intelligence in Health AI in ocular drug discovery and development

The integration of AI into ophthalmic drug discovery to clinical trials, given the substantial costs associated with
represents a convergence of technology and medicine that late-stage drug development failures due to unforeseen
could significantly advance the treatment of eye diseases. By toxicity. AI models, especially those trained on extensive
5
leveraging AI in target identification, compound screening, databases of chemical structures and their associated
and predictive toxicology, researchers can overcome some toxicological profiles, offer high accuracy in predicting
of the traditional bottlenecks in the drug discovery pipeline, the toxicity of new compounds. By integrating data from
5
facilitating much-needed treatments for patients with eye various sources, including preclinical studies and known
diseases. Beyond expediting the development of safe, drug safety profiles, AI can forecast adverse effects and serve
1
effective, and targeted treatments, this integration may also as an early warning system to prioritize compounds with
enrich our understanding of ophthalmic disease processes. favorable safety profiles. This capability not only reduces
6
In this paper, we explore the role of AI in ophthalmic drug the risk of late-stage failures but also ensures a more efficient
discovery, offering insights into its potential to address the allocation of resources toward candidates with the highest
critical challenges in the field. likelihood of success in treating ophthalmic conditions.
It is important to note that while specific applications
2. AI in drug discovery of AI models in ophthalmology are still emerging, the
The discovery and validation of therapeutic targets are theoretical and operational frameworks established in
critical steps in the drug discovery pipeline. This process other therapeutic areas provide a promising foundation for
involves identifying molecular targets, such as proteins their application in ocular drug development.
that play a key role in disease pathogenesis, and confirming
their suitability for therapeutic intervention. Traditional 3. Current applications and case studies in
3
methods for target identification are often labor-intensive, ophthalmology
time-consuming, and fraught with uncertainty. AI can As patients’ adherence to dosing regimens (eye drops) and
analyze vast datasets from genomic, proteomic, and other frequent intraocular injections can be substantial barriers
multi-omic studies to uncover potential targets related to to effective chronic ocular disease management, sustained
eye diseases more efficiently than traditional approaches. drug delivery strategies can be helpful. However, as
3
7
Once potential targets are identified, AI can further assist these sustained therapeutic methods are traditionally
in validating these targets by predicting their role in disease achieved by implantable devices, there is a risk of excipient
progression and response to therapeutic intervention, material buildup, the need for device removal, potential
thus enhancing the specificity and effectiveness of drug adverse reactions, etc. An alternative approach would be
7
development efforts in ophthalmology. 3 to increase the retention time and therapeutic effects of
After identifying possible therapeutic targets, the drugs in the eye without using implants, as attempted by
process of compound screening involves testing numerous Hsueh et al. As ocular melanin has a low turnover rate,
7
agents for activity against the identified target. This phase they hypothesized that a melanin-binding peptide could
also includes optimizing these agents to improve their be conjugated to small-molecule drugs to increase their
efficacy, safety profile, and pharmacokinetic properties. AI retention time and therapeutic effect. Since incorporating
2
has the potential to significantly accelerate this phase using multiple functions into a single peptide sequence is
models that predict compound-target interactions, thereby challenging, they used machine learning methods to help
narrowing down the vast library of potential compounds engineer peptide sequences that could simultaneously
to those most likely to exhibit desired therapeutic effects. provide these desired functions. As a result, their engineered
In addition, AI-driven models can simulate the molecular peptide exhibited increased cell-penetrating properties
docking process and provide predictions on how different and high melanin binding capacity while demonstrating
compounds will bind to target proteins. This not only low cytotoxicity. They tested these compounds in rabbits
4
accelerates the screening process but also enhances the and discovered that their multifunctional peptide greatly
precision of compound selection for further development. enhanced the pharmacokinetics and pharmacodynamics
Moreover, AI can optimize compound structures by of brimonidine when compared to normal use. In this
predicting modifications that enhance drug-like properties, work, machine learning played a key role in identifying
ensuring that the most promising candidates are advanced important variables for desired peptide function, refining
to the next stages of drug development with optimized peptide design, and achieving desired therapeutic goals.
profiles for effectiveness and safety. 2 The application of AI extends beyond enhancing drug
Finally, predictive toxicology is essential in assessing delivery systems to revolutionizing our approach to disease
the safety profile of drug candidates before they advance management strategies, including neurodegenerative

Volume 1 Issue 3 (2024) 27 doi: 10.36922/aih.3341