Multiscalar point and line measurements have provided a wealth of data to extract measured scalars’ statistics in laboratory flames. These measurements are partial, since only a subset of scalars are measured, and carry experimental uncertainty to the degree that derived scalars, such as reaction rates, cannot be adequately evaluated from the raw data alone. In this study, we propose and investigate a method to extract derived scalar statistics via a surrogate composition space. This space is parameterized using principal components (PCs) from principal component analysis (PCA). The resulting low-dimensional manifold based only on statistics from measured scalars is complemented with homogeneous chemistry calculations to recover missing species and evaluate species reaction rates. The method is validated using direct numerical simulation (DNS) data on which Gaussian noise is added to emulate experimental uncertainly and with a subset of major species, a radical OH, and temperature assumed to be known/measured. The results show the proposed procedure is able to recover unmeasured species and predict the species reaction rates.