AI Framework Maps Gene Control Like GPS for DNA

Led by Professor Li Chunquan, the research team built TRAPT to track proteins that regulate genes - a process central to health and disease. The system combines massive datasets with AI to spot which proteins control specific genes.

Think of your genes as light switches. Different cells need different genes "switched on" at different times. A heart cell needs different genes active than a brain cell does. Proteins called transcriptional regulators (TRs) flip these genetic switches.

Finding which TRs control which genes has stumped scientists for years. Earlier tools gave rough guesses - like using a paper map instead of GPS. TRAPT changes this by analyzing over 20,000 datasets showing where TRs bind to DNA.

The system works in two steps. First, it learns where TRs typically bind across all DNA. Then it sees how these patterns change in specific cells or conditions. This two-step approach helps TRAPT make better predictions about gene control.

Tests show TRAPT beats existing tools handily. It spots the right regulatory proteins 13% more often than the next best method. For some types of regulators, it works up to 200% better than current tools.

The team proved TRAPT's real-world value in breast cancer cells. The system found key proteins that drive breast cancer growth, including ESR1 and FOXA1. It also revealed new connections that could help scientists understand the disease better.

TRAPT could help researchers study many diseases beyond cancer. It acts like a universal translator for gene regulation, helping scientists read the complex code cells use to control their genes.

The Shandong team made TRAPT free for scientists worldwide. Researchers can use it through a website or download it to their computers.

"Traditional methods miss a lot of important information about gene control," explains Professor Li. "TRAPT combines AI with comprehensive data to give us a much clearer picture. It's like switching from a paper map to a GPS system for finding gene regulators."

The system especially helps track three types of regulatory proteins:

• Transcription factors that bind directly to DNA
• Transcription co-factors that help other proteins bind
• Chromatin regulators that control DNA packaging

This matters because problems with gene regulation appear in many diseases. Cancer often involves genes turned on when they should be off, or vice versa. Better understanding of gene control could lead to more targeted treatments.

The researchers tested TRAPT on 570 different datasets where scientists had turned specific genes on or off. The system consistently identified the right regulatory proteins, beating other tools across multiple measures.

The team also used TRAPT to study breast cancer, where it correctly spotted proteins known to drive tumor growth. This shows the system can help cancer researchers find important gene regulators.

"We designed TRAPT to handle the complexity of gene regulation," says Professor Li. "It processes huge amounts of data to find patterns that other tools miss."

The system runs on powerful AI techniques including:

• Graph neural networks that map connections between proteins
• Knowledge distillation that helps the system learn efficiently
• Deep learning that spots complex patterns in data

Beyond the tool itself, the team's work provides a valuable resource for other scientists. They collected and organized thousands of datasets about gene regulation, creating one of the field's largest databases.

Why this matters:

• TRAPT gives scientists an unprecedented view of gene control, which could speed up research on cancer, heart disease, and other conditions
• By making both the tool and data freely available, the Shandong team helps researchers worldwide study gene regulation more effectively

Read on, my dear:

• Nature: TRAPT: a multi-stage fused deep learning framework for predicting transcriptional regulators based on large-scale epigenomic data