Structure prediction of protein-ligand complexes from sequence information with Umol

Area of Science:

• Computational Biology
• Drug Discovery
• Artificial Intelligence

Background:

• Protein-ligand docking is crucial for drug discovery but requires high-quality, often rigid, protein structures.
• Existing methods face limitations due to the need for experimentally determined protein structures.

Purpose of the Study:

• To develop an AI system capable of predicting fully flexible, all-atom protein-ligand complex structures directly from amino acid sequences.
• To evaluate the performance of the AI system against classical docking methods.
• To explore the utility of AI-predicted confidence metrics for assessing prediction accuracy and binding affinity.

Main Methods:

• Development of a novel AI system (Umol) for predicting protein-ligand complex structures.
• Utilizing sequence information as input for all-atom, flexible structure prediction.
• Comparison of AI predictions with established classical docking approaches.
• Application of predicted confidence metrics (plDDT) for filtering predictions and distinguishing binder strengths.

Main Results:

• Classical docking methods, when provided with crystal structures, still outperform the developed AI system.
• The AI system successfully predicts fully flexible, all-atom structures of protein-ligand complexes from sequence.
• Predicted confidence metrics (plDDT) effectively aid in selecting accurate predictions and differentiating between strong and weak binders.

Conclusions:

• The developed AI system represents a step forward in AI-driven drug discovery by enabling flexible structure prediction from sequence.
• Further advancements are needed to fully capture the intricacies of protein-ligand interactions using AI.
• The AI system shows promise for guiding experimental drug discovery efforts by providing confidence-based insights.