Page 41 - AIH-2-2

Artificial Intelligence in Health New drug discovery in the AI era

Figure 6 illustrates AI/ML-powered drug discovery and conformational dynamics and multimeric structure
41
development workflow. prediction.
AlphaFold2, for instance, presents new prospects for An improved version, “AlphaFold3,” codeveloped by
structure-based drug discovery, particularly for proteins Google DeepMind and Isomorphic, was released in May
with unverified structures. However, AlphaFold2 data 2024. AlphaFold3 can predict the nature of protein–
39
can be difficult to interpret, and its output in virtual molecule interactions far more effectively than AlphaFold2.
screening has shown inconsistent results when protein While AlphaFold2 is primarily focused on predicting
folding and dynamic conformational changes are not taken protein structures, AlphaFold 3 extends its capabilities to
into account. Although AlphaFold2 performs well in the predict interactions between proteins and a wider range
40
domain of structure prediction, it falls short in addressing of molecules, including DNA, RNA, and small-molecule
ligands. Figure 7 lists several examples of AI-assisted and
generative AI (GenAI)-driven drugs.

3.2. AI-driven drug toxicity prediction
Optimizing the chemical structures of NCEs to enhance
their biological activities while ensuring suitable safety
profiles (such as low in vitro and in vivo toxicity) is
challenging. Newly designed molecules will be less toxic
only if their physicochemical characteristics and potential
off-target effects are considered alongside their biological
activity in a dynamic setting.
The availability of big data and open-access toxicological
datasets has made toxicity prediction feasible. For example,
42
the chemoinformatic platform “ChemTunes•ToxGPS®”
developed by Molecular Networks and Altamira integrates
multiple databases, including physicochemical parameters,
xenobiotic metabolism, toxicokinetics, and the ToxCast/
Tox21 database, to support the safety and risk assessment
of chemical substances. Figure 8 illustrates the role of ML
in drug toxicity prediction.

Figure 6. Artificial intelligence/machine learning-driven drug discovery The discontinuation of the first two AI-designed
workflow clinical candidates (Exscientia’s cancer drug EXS-21546

Figure 7. Examples of artificial intelligence-assisted and GenAI drugs
Abbreviations: A2A: Adenosine A2a receptor antagonist; NSCLC: Non-small cell lung cancer; RCC: Renal cell carcinoma; USFDA: United States Food
and Drug Administration.

Volume 2 Issue 2 (2025) 35 doi: 10.36922/aih.4423