LogP and LogD

LogP is the logarithm of the partition coefficient (P), a measure of the lipophilicity of a neutral compound based on its distribution between octanol and water. Higher logP values generally indicate greater lipophilicity and membrane permeability, while lower values indicate greater hydrophilicity

LogD is the logarithm of the distribution coefficient (D), a pH-dependent measure of lipophilicity describing how a compound partitions between an organic phase (typically octanol) and water. Unlike logP, it accounts for all species present at a given pH, including ionized, partially ionized, and unionized forms. 

In PhysChem Suite, logD is calculated from predicted logP and pK_a values to model the distribution of all species across the two phases at a defined pH, providing a more physiologically relevant measure of lipophilicity. 

LogP describes lipophilicity of neutral compounds only, based on partitioning between octanol and water. LogD is a pH-dependent distribution coefficient that includes both ionized and neutral species. 

LogP reflects intrinsic lipophilicity, while logD reflects behaviour under physiological pH conditions. For more detail, see our blog on when to use logP vs logD.

The logP prediction models are based on datasets containing more than 12,000 experimental logP values and over 22,000 compounds for the GALAS model. Multiple complementary algorithms and advanced similarity-based approaches are used to improve reliability across diverse chemical space.

LogP prediction modules are continuously refined to improve accuracy and expand chemical space coverage. Prediction accuracy depends on chemical class and applicability domain and can be further improved by training supported models with high-quality proprietary experimental data. 
 

Classic Algorithm 

• Based on >12,000 experimental logP values  
• Uses the principle of isolating carbons  
• Provides a mechanistic, fragment-based approach to logP prediction  

GALAS Algorithm 

• Based on a training set of >22,000 compounds  
• Uses similarity-based adjustment from the most similar compounds in the dataset  
• Provides locally refined predictions based on chemical neighbourhood  

Consensus Model 

• Combines Classic and GALAS predictions  
• Applies weighting based on regions of chemical space where each model performs most reliably  
• Provides a balanced prediction leveraging complementary strengths of both approaches 

Predicted logP values may differ from experimentally measured values when: 

• The logP value is outside the reliably measurable range (approximately +8 to -3) obtained by traditional experiments. 
• Tautomeric forms exist. In many cases, logP values cannot be experimentally measured for every tautomer. 
• A compound contains acidic and basic groups (e.g., amino acids, peptides, nucleosides). Since these molecules may exist in multiple ionic forms at a given pH, the concentration of the unionized form may be negligible. 

 In these situations, structure-based predictions may help overcome experimental limitations and provide estimates where experimental measurements become more challenging.