Artificial Intelligence in drug discovery and development: current landscape, challenges, and future perspectives | Shamrock Academic Studio Knowledge Base
Pharmaceutical ScienceAdvanced25 min
Artificial Intelligence in drug discovery and development: current landscape, challenges, and future perspectives
This review examines the transformative impact of artificial intelligence (AI) across the entire drug discovery and development continuum, addressing traditional constraints such as high attrition and immense costs. The authors analyze how generative modeling and large language models (LLMs) enable the systematic exploration of a chemical space exceeding 10^60 molecules to identify novel therapeutic candidates. Key findings reveal that AI-generated drugs demonstrate significantly higher Phase I success rates (80–90%) compared to conventional development (~40%), with the clinical pipeline growing from 3 candidates in 2016 to 67 in 2023. However, the study emphasizes that AI utility is strictly bounded by data quality and the "black box" nature of deep learning architectures. Ultimately, the paper positions AI as a collaborative "lab partner" capable of accelerating target-to-IND timelines while still requiring robust downstream experimental validation.
Key Takeaways
The chemical space for drug discovery is estimated to exceed 10^60 molecules, requiring AI for systematic navigation.
AI-generated drugs achieved an 80–90% success rate in Phase I trials by late 2023, doubling the ~40% rate of conventional drugs.
The number of AI-designed candidates in clinical testing increased from just 3 in 2016 to 67 in 2023.
AI-driven repurposing offers a scalable strategy for over 7,000 rare diseases, fewer than 6% of which have approved therapies.
While the FDA has approved numerous AI-enabled medical devices, no fully AI-designed new molecular entity has received final marketing approval yet.
Learning Objectives
Identify the specific AI platforms used for protein structure prediction, binding site modeling, and scaffold refinement.
Compare the clinical success metrics of AI-originated molecules against traditional medicinal chemistry outcomes.
Explain the methodological challenges of model interpretability, data silos, and benchmarking in pharmaceutical AI.
Distinguish between regulatory frameworks for clinical decision-support software and therapeutic discovery platforms.
Glossary
ADMET
Absorption, Distribution, Metabolism, Excretion, and Toxicity; the criteria used to evaluate a drug's pharmacokinetic and safety profile.
De novo design
The computational generation of novel molecular structures from scratch, rather than modifying existing templates.
Black box problem
A limitation of deep learning where the internal logic and predictive features of a model remain non-interpretable to the user.
Drug repurposing
The process of identifying new therapeutic indications for existing approved or investigational drugs.
QSAR
Quantitative Structure–Activity Relationship; computational models that relate chemical structure to biological activity.
Timeline
2016Only 3 AI-designed drug candidates were in clinical testing.
2018FDA granted breakthrough device status to a Bayer/Merck AI clinical decision support system for CTEPH.
2020DSP-1181 became the first AI-designed small molecule to enter human clinical trials.
2022Baricitinib, an AI-repurposed drug, received full FDA approval for COVID-19 treatment.
2023The number of AI-designed drug candidates in clinical testing reached 67.
2025First patient dosed in trials for REC-3565, an allosteric MALT1 inhibitor designed with AI.
Mind Map
Everything is expanded by default. Use the − buttons to collapse a branch, or the controls below.
AI in Drug Discovery
Target & Hit Discovery
AlphaFold Structure Prediction
MoLeR Scaffold Generation
Preclinical & Clinical Development
DeepTox Toxicity Modeling
Adaptive Trial Design
Major Challenges
Data Silo Limitations
Black Box Interpretability
AI in the Pharmaceutical Pipeline
Transforming Drug R&D Through Computational Excellence
10^60
Molecules in chemical space
80-90%
AI Phase I success rate
67
Clinical candidates (2023)
< 6%
Rare disease therapy access
Timeline Compression
Potential to reduce target-to-IND timelines from 7 years to just 2-3 years.
Human-AI Synergy
AI acts as a 'lab partner' uncovering designs humanly impossible to parse.
Has any fully AI-designed drug received final FDA marketing approval?
No fully AI-discovered and AI-designed drug has yet received marketing approval, although several candidates are in Phase II clinical trials.
What is the difference between AI-assisted and AI-originated drugs?
AI-assisted drugs use machine learning to optimize existing human-derived scaffolds, while AI-originated drugs are generated de novo by autonomous models without a human template.
How does AI impact rare disease research?
AI provides a systematic, data-driven strategy for drug repurposing, which is vital for the 7,000+ rare diseases where small patient populations hinder traditional R&D.
What is the role of AlphaFold in drug discovery?
AlphaFold predicts highly accurate 3D protein structures from amino acid sequences, establishing the structural topology needed to model drug-binding sites.