AI in Drug Discovery: Hype vs Reality

Artificial intelligence has captured the imagination of the pharmaceutical industry. Headlines proclaim AI will revolutionise drug discovery, compressing timelines from years to months and dramatically reducing the cost of bringing new medicines to market. But as investors, we must look beyond the hype to evaluate what AI can actually deliver today and where the real opportunities lie.

The Data: Early Signs of Promise

According to a rigorous analysis published in Drug Discovery Today, researchers examined the clinical pipelines of AI-native biotech companies to assess whether AI-discovered molecules perform differently in human trials. The findings are illuminating.

In Phase I trials, AI-discovered molecules demonstrated an 80-90% success rate, substantially higher than historic industry averages of approximately 60%. This suggests AI is highly capable of designing or identifying molecules with drug-like properties, optimising for the pharmacokinetic and safety parameters that determine whether a compound can safely advance in humans.

Where AI Falls Short (For Now)

Phase II success rates tell a more nuanced story. At approximately 40%, AI-discovered compounds perform comparably to historic industry averages. This reveals an important truth: AI has not yet cracked the fundamental challenge of predicting whether a drug will work in patients.

As researchers from Cambridge and the European Bioinformatics Institute note, changes in clinical success rates will have the most profound impact on improving drug discovery outcomes. The quality of decisions regarding which compound to take forward remains more important than speed or cost optimisation.

Current AI approaches excel at answering "how to make a compound" but struggle with the more fundamental question of "which compound to make." The field's focus on optimising synthesis routes, binding affinity, and ADMET properties, while valuable, addresses only part of the drug discovery equation.

The Data Challenge

A critical bottleneck for AI in drug discovery is data quality and relevance. Current proxy measures and available datasets cannot fully utilise AI's potential, particularly for predicting drug efficacy and safety in vivo. The industry's reliance on preclinical models that often fail to translate to human outcomes limits what any algorithm can learn.

Addressing which data to generate and which endpoints to model will be key to improving clinically relevant decision-making. This represents a significant opportunity for companies that can generate proprietary, high-quality datasets linking molecular features to clinical outcomes.

The Interpretability Imperative

Beyond raw predictive power, ensuring the interpretability of AI models is critical for securing regulatory approval and building trust within scientific and medical communities. Black-box algorithms that cannot explain their predictions face significant hurdles in a regulated industry where mechanistic understanding informs risk-benefit decisions.

Our Investment Framework

Based on our analysis, we've developed a nuanced framework for evaluating AI-drug discovery companies:

Tier 1: Molecular Optimisation
AI demonstrably accelerates lead optimisation, improving compound properties and reducing cycle times. Companies with validated platforms in this space offer near-term value creation.

Tier 2: Target Identification
AI can help identify novel targets by mining genomic, proteomic, and clinical data. The value here depends heavily on proprietary data access and validation capabilities.

Tier 3: Clinical Prediction
The holy grail: predicting which compounds will work in humans. This remains largely aspirational, but companies making genuine progress here will capture outsized value.

The Path Forward

We remain optimistic about AI's transformative potential in drug discovery, while maintaining discipline about current capabilities. The technology is genuinely useful today for specific applications, and rapid advances in foundation models, multi-modal learning, and biological data generation suggest meaningful improvements ahead.

For investors, the key is distinguishing between companies that have deployed AI for incremental efficiency gains versus those building towards the harder problem of predicting clinical success. Both can create value, but the latter represents the larger opportunity.