Multi-target QSPR assemble of a Complex Network for the distribution of chemicals to biphasic systems and biological tissues

Chemometrics, that based prediction on the probability of chemical distribution to different systems, is highly important for physicochemical, environmental, and life sciences. However, the amount of information is huge and difficult to analyze. A multi-system partition Complex Network (MSP-CN) may be very useful in this sense. We define MSP-CNs as large graphs composed by nodes (chemicals) interconnected by arcs if a pair of chemicals have similar partition in a given system. Experimental quantification of partition in many systems is expensive, so we can use a Quantitative Structure–Partition Relationship (QSPR) model. Unfortunately, with classic QSPR we need to use one model for each system. Here construct the first MSP-CN based on a multi-target QSPR (mt-QSPR). The model is based on the spectral moments (πk) of a molecular Markov matrix weighted with atomic parameters that depend on both the nature of the atom and the partition system. The mt-QSPR predicts 90.6% of 413 compound/system pairs in training series and 90.0% in validation. The MSP-CN predicted presents 413 nodes, 2060 edges, average node degree 9.9, and only 7.7% drugs are unconnected. The model was used to study the biophysical phenomena of transport or distribution of G1 (a novel antimicrobial drug) to different rat tissues. Predicted probabilities (P) coincide with low experimental partition coefficients (logPC) reported herein by the first time in skin (P = 0.455; logPC = − 0.02 < 0 → U), heart (0.453; − 0.02 → U), and brain (0.324; − 0.34 → U). The Kamada–Kawai algorithm evidenced the community structure of the MSP-CN and clusters G1 into three different communities of the U-type drugs. These results coincide with the low distribution of G1 to these tissues and consequently have low expected drug side effect.