Comparability of mixed IC₅₀ data - a statistical analysis

Area of Science:

• Biochemistry
• Cheminformatics
• Pharmacology

Background:

• Biochemical half maximal inhibitory concentration (IC50) is crucial for drug discovery, guiding lead optimization and predictive modeling.
• Public IC50 data are widely used but suffer from assay specificity, limiting comparability and large-scale analysis.
• Manual validation of individual data entries is infeasible for extensive datasets.

Purpose of the Study:

• To analyze errors, redundancy, and variability in the ChEMBL IC50 database.
• To assess the impact of mixing public IC50 data from different assays.
• To investigate the feasibility of augmenting IC50 data with inhibition constant (Ki) data.

Main Methods:

• Analyzed error types, redundancy, and variability within the ChEMBL IC50 database.
• Assessed IC50 data variability by comparing independent measurements from different labs and all pairwise comparisons.
• Evaluated the effect of augmenting mixed IC50 data with corrected public Ki data.

Main Results:

• The standard deviation of IC50 data is only 25% larger than that of Ki data, indicating moderate noise increase when mixing IC50 values.
• Public IC50 data variability is higher than in-house intra-laboratory data, as expected.
• Augmenting mixed public IC50 data with public Ki data, using a Ki-IC50 conversion factor of 2, does not degrade data quality.

Conclusions:

• Mixing public IC50 data, despite assay variability, introduces only a moderate amount of noise.
• Public IC50 data can be effectively augmented with public Ki data using a conversion factor of 2.
• This approach enhances the utility of public databases for large-scale chemogenomics and predictive modeling.