Systematic evaluation of connectivity map for disease indications

Area of Science:

• Genomics
• Pharmacology
• Bioinformatics

Background:

• Connectivity Map (CMAP) data aids in understanding drug mechanism of action (MOA) and discovering new drug indications.
• CMAP's core principle involves measuring connectivity between disease gene expression signatures and compound-induced profiles.
• Previous evaluations of CMAP's accuracy in drug-indication prediction have been limited, often relying on indirect drug-to-drug matching.

Purpose of the Study:

• To directly assess CMAP methodologies for classifying known drug-disease relationships.
• To evaluate the prediction performance of three CMAP-based methods using a curated dataset.

Main Methods:

• Evaluated three CMAP-based methods on a dataset of 890 true drug-indication pairs.
• Generated disease signatures using the Gene Logic BioExpress™ system.
• Derived compound profiles from the Connectivity Map database (CMAP, build 02).

Main Results:

• The eXtreme Sum (XSum) similarity scoring algorithm outperformed the standard Kolmogorov-Smirnov (KS) statistic.
• XSum achieved a four-fold enrichment at a 0.01 false positive rate.
• Statistical significance was demonstrated with AUC = 2.2E-4 and P value = 0.0035.

Conclusions:

• Connectivity Map analysis can significantly enrich true positive drug-indication pairs.
• The effectiveness of CMAP in this context is dependent on employing an efficient matching algorithm like XSum.