From Big Data to Big Brains: the Path to Smarter Drug Design in Canada
June 1, 2025

From Big Data to Big Brains: the Path to Smarter Drug Design in Canada


EVERY DAY, WE HEAR about how artificial intelligence is ready to revolutionize Canadian life sciences. Claims that AI can or will make most things ‘faster, cheaper, better’ are already ubiquitous. And while we are seeing AI’s power in being able to analyze vast datasets—in minutes, not months—to pinpoint potential drug candidates or disease markers, there’s an important question we need to ask: is AI intelligent enough to deliver on all that’s being promised?

THE LIMITS OF LEARNING BY EXPOSURE

Many current AI approaches, particularly in machine learning, are designed to learn from exposure to the experience contained in vast amounts of data. Feed an algorithm enough examples of successful drugs, failed drugs, molecular structures, and bioactivity data, and it becomes adept at spotting correlations and predicting outcomes for similar new inputs.

But drug discovery is inherently messy and complex. Biological systems are intricate; the path from molecule to medicine is rarely linear; and the data that scientists are working with is often sparse, heterogeneous, and biased. So, relying solely on trends observed in historical data has significant drawbacks:

  1. “Black Box” problem: Machine learning models can predict outcomes but not explain them, making their predictions hard to trust, especially in new chemical or biological scenarios.
  2. Sensitivity to data bias and the struggle with novelty: AI tends to perpetuate biases in historical data, potentially overlooking innovative solutions outside familiar patterns. And, while adept at using existing data, AI often falters in creating novel molecular designs or addressing unprecedented biological challenges.
  3. Lack of foundational understanding: AI doesn’t truly understand scientific principles, recognizing functional group interactions based only on data patterns, not underlying chemistry.

In all the cases above, AI isn’t giving us true intelligence in the way a human scientist can. Instead, it’s sophisticated mimicry based on statistical correlation—powerful, sure, but fundamentally limited.

REDEFINING AI INTELLIGENCE

Imagine giving an 11-year-old a pile of data, expecting them to understand it and make recommendations to help you make a drug? That kid represents AI in drug design today: able to see patterns but lacking the intelligence to make sense of it. At the Conscience Symposium on Open Drug Discovery in Montreal earlier this year, renowned medicinal chemist and blogger, Derek Lowe, framed this as AI being able to give us answers, but not the reasoning behind them.

It’s time to stop showing AI millions of examples of things, hoping it intuits rules. To unlock the next level of AI-driven innovation in drug design, AI needs genuine “intelligence,” rooted in the fundamental meaning of the word: the ability to learn, understand, and apply knowledge.

TEACHING FOUNDATIONAL PRINCIPLES

Before asking an AI to design a drug, we need to teach it the basics, much like a human student, and make it smarter. This doesn’t mean feeding the AI raw data points that represent the principles of chemistry, physics, and biology; it means embedding the principles themselves into the AI’s architecture or knowledge base.

Think of it like giving the AI the textbooks and making sure it understands the core concepts before it sees specific experimental results. The goal is an AI that possesses a rudimentary, rule-based understanding of how the world works at a molecular and biological level.

DEVELOPING CRITICAL THINKING AND LEARNING HOW TO SOLVE PROBLEMS

Once the AI has foundational scientific understanding, we need to teach it to apply these principles, integrating its foundational knowledge with real-world experimental data. Here, the AI learns to:

  • Reason, moving from correlation to understanding causation based on scientific mechanisms.
  • Hypothesize, generating molecular ideas grounded in chemical and biological plausibility.
  • Critically evaluate, assessing drug candidates with a comprehensive understanding of their potential benefits and risks.
  • Adapt, using fundamental rules to handle sparse or conflicting data for more robust predictions.

Once it’s learned these things, the AI can move beyond replicating patterns and instead can engage in computational reasoning. For example, it can make smarter decisions in computational chemistry because it understands the underlying science. It can suggest genuinely novel molecules because its creativity isn’t constrained solely by the limits of the training data.

IMPLICATIONS FOR CANADIAN BIOTECH

Canada’s biotech sector is ready for truly intelligent AI in the creative and complex field of drug design.

Intelligent AI could help us design novel chemical entities for challenging targets, predicting drug properties with higher accuracy and interpretability, and reducing late-stage failures by identifying issues earlier based on mechanistic understanding. It could help us develop tools that interpret complex biomarker signatures based on an understanding of underlying disease biology, not just statistical patterns. And imagine the promise of AI that could create tailored therapeutic strategies by simulating how drug candidates might interact with an individual’s specific biological makeup, informed by both genomic data and foundational biological principles.

There is no better time to support Canadian companies developing educated AI systems that truly understand science. It’s also time to move beyond the hype that AI will make things faster, cheaper, better, and instead focus on turning AI into an educated partner that can help us develop smarter, more innovative solutions. We have a shared opportunity and responsibility to work together, now more than ever, to grow Canada’s life sciences ecosystem. This is about changing the game completely to position Canada at the forefront of a new era in computationally driven biomedical discovery, ready to tackle the toughest challenges in human health.

Josh Pottel is CEO at Molecular Forecaster Inc. (MFI), where he and his team have been teaching chemistry to their proprietary software for years, helping organizations make smarter decisions in drug design. MFI’s goal? To become the go-to partner in small molecule drug design, combining proprietary tools and technology, deep expertise and know-how, and an all-in approach to collaboration that sets MFI’s partners up for success.