Cell-specific imputation of drug connectivity mapping with incomplete data.

Area of Science:

• Computational biology
• Pharmacology
• Genomics

Background:

• Drug repositioning accelerates the discovery of new uses for existing drugs, but screening large compound libraries is costly.
• Connectivity mapping links drugs to diseases by identifying compounds that reverse disease-associated gene expression changes.
• The LINCS project has increased available compound and cell data, yet many useful drug combinations remain uncharacterized.

Purpose of the Study:

• To evaluate computational methods for predicting drug connectivity and enabling drug repositioning despite missing experimental data.
• To compare the performance of collaborative filtering (neighborhood-based and SVD imputation) against naive approaches in predicting drug-disease relationships.

Main Methods:

• Cross-validation was used to assess the predictive accuracy of different imputation methods for drug connectivity.
• Collaborative filtering techniques, including neighborhood-based and Singular Value Decomposition (SVD) imputation, were compared.
• The influence of cell type on prediction accuracy was investigated.

Main Results:

• Collaborative filtering methods significantly improved the prediction of drug connectivity compared to naive approaches, especially when cell type was considered.
• Neighborhood collaborative filtering demonstrated the highest success rate, particularly in non-immortalized primary cells.
• The study identified compound classes that are more or less dependent on cell type for accurate imputation.

Conclusions:

• Computational methods, particularly neighborhood collaborative filtering, can successfully predict drug responses and identify potential drug repositioning candidates even when experimental data is incomplete.
• Cell type is a crucial factor in improving the accuracy of drug effect prediction.
• This approach facilitates the identification of unassayed drugs that can reverse disease-specific expression signatures in various cell types.