Extracting functional insights from loss-of-function screens using deep link prediction.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Loss-of-function screens are crucial for identifying genes essential for cell survival.
• Interpreting complex screening data and prioritizing candidate genes remains a challenge.

Purpose of the Study:

• To introduce Deep Link Prediction (DLP), a novel method for interpreting loss-of-function screening data.
• To enhance the identification and prioritization of essential genes and potential drug targets.

Main Methods:

• DLP utilizes representation-based link prediction to integrate screening experiments with gene-gene interaction networks.
• The method was validated using RNA interference (RNAi) and CRISPR screening datasets from DepMap across multiple cancer types.

Main Results:

• DLP-DeepWalk significantly outperformed existing methods in identifying cell-specific dependencies, achieving over 90% average precision.
• Genes prioritized by DLP-DeepWalk showed higher enrichment in known drug targets compared to original screening scores.
• The enrichment of drug targets was more pronounced in RNAi data than CRISPR data, reflecting differences in data noise.

Conclusions:

• DLP offers a powerful approach for re-prioritizing phenotypic readouts from loss-of-function screens.
• DLP enhances the discovery of therapeutically relevant genes and provides insights into their mechanisms of action.
• The method's effectiveness across different screening technologies and cancer types highlights its broad applicability.