Lateral heterogeneity, or mosaicity, is a fundamental property inherent to cell membranes that is crucial for their functioning. While microscopic inhomogeneities (e.g. rafts) are easily detected experimentally, lipid domains with nanoscale dimensions (nanoclusters of nanodomains, NDs) resist reliable characterization by instrumental methods. In such a case, important insight can be gained via computer modeling. Here, NDs composed of lipid's head groups in the mixed zwitterionic dioleoylphosphatidylcholine (DOPC) and negatively charged dioleoylphosphatidylserine (DOPS) bilayers were studied by molecular dynamics. A new algorithm has been developed to identify NDs. Unlike most similar methods, it implicitly considers the heterogeneous distribution of lipid head atomic density and does not require subjectively chosen parameters. In DOPS-rich membranes, lipids form more compact and stable NDs due to strong interlipid interactions. In DOPC-rich systems, NDs arise due to the "packing" effect of weakly bound lipid heads. The clustering picture is related to the physical properties of the bilayer surface: DOPS-rich systems show more pronounced surface heterogeneity of hydrophilic/hydrophobic regions compared to DOPC-rich ones. The results obtained are important for the effective quantitative characterization of the "dynamic molecular portrait" of a membrane surface - its "fingerprint" characterizing dynamical distribution of its physicochemical properties.