Page 34 - AIH-1-3

Artificial Intelligence in Health AI in ocular drug discovery and development

ocular diseases. There is a plethora of neurodegenerative algorithms. These models are often described as “black
conditions that can cause damage to the optic nerve. One boxes” due to their opaque decision-making processes,
8
of the primary pathophysiological mechanisms of action which are difficult for humans to comprehend. This
11
8
involves the damage of retinal ganglion cell (RGC) axons. lack of transparency can hinder the trust and acceptance
There is emerging evidence suggesting that axonal damage of AI-driven discoveries among clinicians, researchers,
can initiate RGC death through reactive oxygen species and regulatory bodies, which is critical for translating
(ROS), which in turn increases disulfide bond formation AI discoveries into practical therapeutic interventions.
between cysteine side chains to cause further cellular To address these challenges, the current strategies focus
damage. Redox-active phosphine-borane complexes have on the development of explainable AI methods, such as
8
been proposed as protective molecules that can activate feature importance scores and rule-based decision trees,
cellular pathways to prevent these disulfide bonds from designed to demystify AI decisions and enhance model
forming. However, limited pharmacological data exists transparency. In addition, integrating domain-specific
8
for these compounds. To resolve this issue, Remtulla knowledge and employing hybrid models that combine
et al. trained neural networks on features such as cellular deep learning with interpretable statistical methods are
8
permeability, oral absorption, blood–brain barrier proving crucial in improving both the interpretability
permeability, and serum protein binding to reliably predict and reliability of these systems. Ensuring robustness
the pharmacokinetics of boron-containing compounds. and generalization through rigorous testing, coupled
Their results revealed that phosphine-boron compounds with proactive stakeholder engagement, is essential to
met the necessary pharmacokinetic profile to function validating and gaining acceptance for AI technologies in
as orally active drug candidates. Ultimately, this study clinical settings.
underscores the innovative use of machine learning in The integration of AI into drug discovery also
evaluating the pharmacokinetics of emerging compounds, presents ethical and regulatory challenges. The use of
such as phosphine-borane complexes, advancing their patients’ data raises privacy concerns, requiring stringent
potential as neuroprotective agents against RGC damage. data protection measures and ethical oversight to
It exemplifies the ability to generate new perspectives ensure patient confidentiality and consent. Moreover,
3
in ocular pharmacology using pre-existing data and AI regulatory frameworks for AI-assisted drug discovery
algorithms. and development are still in their infancy, lacking clear

4. Challenges and limitations guidelines for validation, approval, and oversight of
AI-driven methodologies. This regulatory uncertainty can
Despite the promising advancements and successful delay the adoption and application of AI technologies in
applications of AI in ophthalmologic drug discovery, ophthalmology drug discovery.
several challenges and limitations remain that require
acknowledgment and resolution. First and foremost, 5. Future direction
the quality and quantity of data available for AI models The integration of AI into ophthalmologic drug discovery
significantly influence their performance and reliability. marks a new era of medical innovation and operational
9
In the realm of ophthalmology, high-quality, diverse, efficiency, addressing longstanding challenges and opening
and annotated datasets, especially from clinical settings, new avenues for therapeutic development. Looking ahead,
are often scarce or fragmented. This limitation can lead several research objectives are set to further leverage the
9
to biases in AI models, reducing their generalizability capabilities of AI systems. Among these, a key goal will
and accuracy when applied to broader, more diverse involve employing these technologies to enhance our
populations. understanding of complex eye diseases at the molecular
Furthermore, the computational resources required for level. Future efforts are likely to focus on developing more
AI research are substantial. The processing of large datasets sophisticated algorithms that can process and analyze
and the training of sophisticated models necessitate advanced the increasingly large and complex datasets generated
hardware and significant computational power, which can by biomedical research. This will not only improve the
9
be a barrier for institutions with limited resources. This accuracy of target identification and validation but also
10
technological and financial barrier may lead to disparities in enable the discovery of novel biomarkers and therapeutic
research advancements and the adoption of AI technologies targets. 2
across different regions and institutions.
Another promising direction involves the integration of
Another significant challenge is the interpretability AI with other emerging technologies, such as gene editing
of AI models, particularly those based on deep learning and stem cell therapy. By combining AI’s predictive
12

Volume 1 Issue 3 (2024) 28 doi: 10.36922/aih.3341