AI is transforming drug discovery, but not in the way many imagine. Rather than algorithms working alone, the most promising advances combine machine learning with physics-based molecular simulations, mechanistic models, and rigorous laboratory validation. This pragmatic approach is reshaping how researchers develop safer, more effective therapies.

Why AI Alone Isn't Enough for Drug Discovery?

Michael Antonov, who pioneered virtual reality as co-founder of Oculus, has spent recent years exploring computational drug discovery. His perspective on AI in medicine cuts through the hype. "There is still a lot of uncertainty," he explained, noting that the pharmaceutical industry faces a challenge: hundreds of companies and projects claim to use AI, but distinguishing genuinely useful tools from overhyped ones remains difficult.

The core problem is that AI models, no matter how sophisticated, cannot solve fundamental challenges alone. Toxicity prediction remains one of the most stubborn obstacles in drug development. While dozens of AI-powered toxicity predictors exist, Antonov noted that "they're not necessarily super great, though they're better than nothing." The real test, he emphasized, comes in the laboratory.

Toxicity

"The biggest thing is to see what you actually prove in the lab," said Michael Antonov, co-founder of Deep Origin.

Michael Antonov, Co-founder of Deep Origin

How Is Deep Origin Combining AI With Physics?

Antonov co-founded Deep Origin to integrate machine learning with physics-based molecular simulations and mechanistic models. This hybrid approach aims to improve virtual drug screening, predict toxicity more accurately, and help researchers develop safer compounds. Rather than relying on pattern recognition alone, the platform uses computational physics to understand how molecules actually behave.

The philosophy behind this strategy reflects a broader shift in biomedical research. Many researchers recognize that AI excels at finding patterns in data, but understanding the underlying biology requires deeper mechanistic insight. By combining both approaches, Deep Origin's AI-native platform can support the full discovery process for small molecule drugs with more predictable outcomes.

What Could Personalized Medicine Look Like in the Next Decade?

Antonov envisions a future where AI-powered drug discovery becomes truly personalized. Within 20 years, he suggested, the process could work like this: a patient with a novel genomic mutation or a new viral infection provides their data to a system that analyzes it comprehensively. The system would then decompose the disease pathways, identify specific drug targets, and recommend the exact concentrations needed for that individual.

This vision depends on several technological advances working together. Current drug development fails roughly 90 percent of the time; Antonov believes AI-driven toxicity models could reduce that failure rate to around 60 percent, a massive improvement. Physics-based counter-screening can computationally identify molecules that should not bind to off-target proteins, reducing unwanted side effects.

By 2030, Antonov predicted, the integration of genomic data with population-level insights could make medicine far more individualized. This requires detailed mechanistic models that go beyond simple pattern matching. The convergence of AI, physics, and genomics could enable researchers to move from 90-month timelines to just a few months for identifying good drug candidates.

How to Evaluate AI Tools in Drug Discovery

• Laboratory Validation: Demand proof that AI predictions hold up in actual experiments, not just in computational models or benchmark tests.
• Mechanistic Understanding: Look for tools that combine machine learning with physics-based simulations and biological models, not AI alone.
• Toxicity Prediction Accuracy: Assess whether the tool's toxicity models are trained on both experimental data and computational physics, since this combination produces more reliable predictions.
• Transparency on Limitations: Be skeptical of vendors claiming AI solves all problems; the best tools acknowledge what they cannot do and where human expertise remains essential.

What Is the "AI Scientist" Trend in Drug Discovery?

One emerging trend gaining momentum is the concept of an "AI scientist" that can autonomously identify drug targets and design molecules. Deep Origin received a grant from the United Kingdom to test this approach on endometriosis, a disease that affects millions of people worldwide. The AI system analyzed hundreds of potential drug targets using specific scientific criteria, then selected the most promising one with detailed reasoning.

Antonov believes this automation could democratize drug discovery. If AI systems can handle the routine steps that have been performed many times before, researchers and even patient advocacy groups could take significant steps toward developing new therapies without requiring massive pharmaceutical infrastructure. However, he stressed that human scientists remain critical to validating results and making strategic decisions.

"It doesn't mean that AI will do all the steps for us, but it can do a lot of the known steps, which have been done many times and can help us along the way. Of course, the real scientist will still be very critical to all the parts," Antonov noted.

Michael Antonov, Co-founder of Deep Origin

How Does This Compare to Other AI-Driven Cancer Detection Advances?

While drug discovery focuses on developing new therapies, another major breakthrough is reshaping early cancer detection. Caris Life Sciences launched Caris Detect, a multi-cancer early detection blood test that combines ultra-deep whole genome sequencing, whole transcriptome sequencing, and artificial intelligence algorithms.

Caris Detect analyzes a comprehensive spectrum of molecular and biological signals, including DNA, RNA, and other patterns, rather than relying on limited genomic snapshots. The test is powered by Caris' molecular profiling database, which includes data from over one million cases and more than 50 billion molecular markers. This massive dataset enables AI models to detect subtle signals associated with early-stage disease across multiple cancer types.

"Early detection remains one of the most urgent and consequential challenges in oncology. Caris Detect represents a major step forward in our mission to transform cancer care through the most advanced molecular science available," said David Dean Halbert, Founder, Chairman and Chief Executive Officer of Caris Life Sciences.

David Dean Halbert, Founder, Chairman and Chief Executive Officer of Caris Life Sciences

The launch builds on data from the Caris Detect ACHIEVE 1 study, which demonstrated strong performance in detecting cancer signals across multiple cancer types. Caris intends for Caris Detect to replace older screening methods as the new standard of care for early cancer detection.

Both Deep Origin's approach to drug discovery and Caris Life Sciences' cancer detection platform reflect a shared principle: combining AI with comprehensive biological data and rigorous validation produces better outcomes than either approach alone. As these technologies mature, the convergence of machine learning, genomics, and mechanistic understanding will likely define the next decade of precision medicine.