Page 38 - AIH-2-2

Artificial Intelligence in Health New drug discovery in the AI era

Figure 2. Biotransformation of the peroxisome proliferator-activated receptor gamma agonist “Troglitazone”

including rodent PK, efficacy testing in a rodent disease and human responses due to species differences. These
16
model, and acute toxicity experiments. If the compound translational limitations have heightened concerns that
proved effective and was expected to achieve the same animal studies may mislead us, contributing to clinical
efficacy in humans at a total daily dose of ≤100 mg with candidate failures. It is also important to note that
manageable toxicity, the pre-clinical screening approach selecting the appropriate non-rodent species for pre-
for backup/follow-on compounds was reversed. New clinical evaluation plays a key role in the clinical success of
variants of that scaffold were first subjected to in vitro NCEs. For example, in evaluating a new oral drug with an
profiling, followed by a mouse cassette PK study. extended or sustained-release formulation, minipigs may
provide more relevant data than beagle dogs. This is because
The in vivo toxicity of each variant was assessed, pigs more closely resemble humans than beagle dogs in
followed by efficacy studies. Widening of the therapeutic terms of gut anatomy (intestinal length per kilogram of
index allows for the acceptance of somewhat lower potency. body weight), physiology, bacterial gut colonization, skin

Figure 3 illustrates the in vitro and in vivo tests specified architecture, and body fat distribution. 17
by the Organization for Economic Co-operation and
Development to assess the genotoxicity, mutagenicity, and 2.5. Animal testing alternatives
organ-specific toxicity of NCEs as well as the corresponding The USFDA Modernization Act 2.0 allows for alternatives
AI/machine learning (ML) tools for in silico prediction. to animal testing, enabling the use of pre-clinical assays
that utilize organ-on-chip platforms, organoids, and
2.4. Drug toxicity effect translation from animals to 3D spheroid cultures of human origin to better predict
humans potential toxicities in humans. Recently, the USFDA
However, toxicological data derived from animal models favorably considered pre-clinical efficacy results from
align with human outcomes in only 63% of cases when human organ-on-chip research, alongside existing safety
extrapolated from non-rodents and 43% from rodents and data, to approve the clinical trial IND application of
< 30% when predicting adverse drug reactions in humans. sutimlimab developed by Sanofi.
In addition, drug-induced neurobehavioral symptoms such Notably, primary cultures of cells derived from the human
as nausea, somnolence, and dizziness, which are common heart, liver, or kidney can display differentiated functions, in
in patients and often lead to intolerance, are poorly addition to toxicity markers, and mimic responses observed
predicted by animal studies using conventional endpoints. in intact tissues. However, the challenge of maintaining a
15
Approximately 90% of drug candidates fail, largely because normal gene expression profile and dynamic biochemical
animal studies cannot reliably predict efficacy, safety, responses to varying drug concentrations remains unresolved.

Volume 2 Issue 2 (2025) 32 doi: 10.36922/aih.4423