EXP2SL: A Machine Learning Framework for Cell-Line-Specific Synthetic Lethality Prediction.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Synthetic lethality (SL) is a key genetic interaction for identifying cancer therapeutic targets.
• Most known SL gene pairs are cell-line specific, limiting broad applicability.
• Gene expression profiles offer potential features for predicting SL interactions.

Purpose of the Study:

• To develop a computational method for predicting human synthetic lethality interactions.
• To leverage gene expression profiles from the LINCS L1000 dataset for SL prediction.
• To evaluate the performance of the developed method against existing approaches.

Main Methods:

• Developed EXP2SL, a semi-supervised neural network model.
• Utilized gene expression profiles from shRNA perturbation experiments (LINCS L1000).
• Systematically evaluated the model on SL datasets from three distinct cell lines.

Main Results:

• EXP2SL accurately identified synthetic lethality interactions.
• The model outperformed baseline methods in prediction accuracy.
• Demonstrated the effectiveness of L1000 gene expression features and semi-supervised learning.

Conclusions:

• Gene expression profiles from LINCS L1000 are valuable for predicting human synthetic lethality.
• EXP2SL provides an effective approach for identifying SL interactions.
• This method aids in the discovery of novel anticancer therapeutic targets.