A multi-layer encoder prediction model for individual sample specific gene combination effect (MLEC-iGeneCombo).

Area of Science:

• Systems biology
• Genomics
• Computational biology

Background:

• Synthetic lethal (SL) gene pair identification is crucial for targeted cancer therapies.
• Existing SL prediction models show high inconsistency due to reliance on various SL scores.
• A reliable measurement for gene combination effects is needed.

Purpose of the Study:

• Introduce a novel, consistent gene combination effect (GCE) measurement.
• Develop a deep learning model (MLEC-iGeneCombo) for sample-specific GCE prediction.
• Enable prediction of GCE for previously unseen cell lines.

Main Methods:

• Developed a new GCE measurement: log-fold change of dual-gRNA expression post-CRISPR-cas9 transfection.
• Constructed MLEC-iGeneCombo, a multi-layer encoder model incorporating sample-specific multi-omics, network, and cell-line information.
• Utilized data from 18 gene combination double knockout (CDKO) experiments.

Main Results:

• The new GCE measurement demonstrated high consistency across CDKO experiments.
• MLEC-iGeneCombo achieved an average GCE prediction performance of 71.9% on 18 CDKO experiments.
• All three encoders (multi-omics, network, cell-line) significantly improved prediction accuracy, with their combination yielding the best results.

Conclusions:

• The novel GCE measurement provides a direct and consistent way to assess gene combination effects.
• MLEC-iGeneCombo represents a significant advancement in predicting GCE, offering sample-specific predictions.
• The multi-layer encoder approach enhances prediction accuracy, paving the way for more reliable SL gene pair discovery.