AI Tool Finds Hidden Drug Targets Across Cancers

A powerful new artificial intelligence model, DeepTarget, is reshaping the way scientists understand cancer drugs, revealing that many medicines may have untapped uses in tumor types their developers never intended (1^✔ ✔Trusted Source
DeepTarget predicts anti-cancer mechanisms of action of small molecules by integrating drug and genetic screens

Go to source). Developed at Sanford Burnham Prebys and published in npj Precision Oncology, DeepTarget challenges a long-standing assumption in drug development: that small-molecule drugs act primarily through one intended target, and everything else is merely a side effect.

Instead, the tool shows that many drugs behave very differently depending on the cancer type, genetic makeup, and cellular context, creating opportunities to repurpose existing drugs for entirely new patient groups.

How DeepTarget Works

Unlike traditional approaches that rely on chemical structures or predicted docking interactions, DeepTarget integrates real experimental data from large-scale cancer drug screens and genetic dependency maps.

Its training dataset included:

• 1,450 drugs
• 371 cancer cell lines
• Comprehensive DepMap functional screens

When benchmarked against leading tools like RoseTTAFold All-Atom and Chai-1, DeepTarget consistently delivered more accurate predictions of drug targets in seven of eight trials.

Crucially, the AI system could also:

• Distinguish whether cancers were sensitive to the normal or mutant form of a protein
• Identify secondary drug targets that may drive therapeutic effects
• Explain why some tumors respond to drugs even when the expected target is absent

Case Study: Why Ibrutinib Works in Lung Cancer

Ibrutinib is FDA-approved for blood cancers targeting BTK. Yet clinical reports showed that some lung cancer patients also respond, despite their tumors lacking BTK altogether. DeepTarget predicted that in lung tumors, the drug was acting instead on a mutant form of EGFR, a major oncogenic driver in lung cancer.

Laboratory validation confirmed:

• Lung cancer cells with mutant EGFR were far more sensitive to Ibrutinib
• Cells without the mutation showed minimal response

This demonstrated a clear context-specific target shift and highlighted how secondary targets can explain unexpected clinical benefits.

A New Framework for Drug Repurposing

Many cancer drugs approved today have multiple targets, but these interactions are often ignored or mislabeled as undesirable. DeepTarget reframes them as opportunities.

The model successfully predicted secondary targets in dozens of well-characterized cancer drugs, opening the possibility of re-examining entire drug libraries for hidden therapeutic potential. Researchers say this approach mirrors real biological behavior more closely than molecular docking or structural predictions, which often miss pathway-level interactions.

What Comes Next

The team aims to expand DeepTarget into a platform for:

• Systematically repurposing FDA-approved drugs
• Designing new small molecules with multi-target potential
• Mapping cancer vulnerabilities that cannot be detected through traditional methods

With millions of chemical compounds still unexplored, AI-driven biological prediction may be essential for accelerating future cancer drug discovery.

Reference:

1. DeepTarget predicts anti-cancer mechanisms of action of small molecules by integrating drug and genetic screens - (https://www.nature.com/articles/s41698-025-01111-4)

Source-Medindia