v2.0

• Message passing neural network architecture, that is capable of representing arbitrary binary solvent mixtures with different mixture ratios and temperatures.
• Absolute solubilities, a absolute-solubility MPNN is trained along the differential version. Then, instead of the mean procedure outlined in the preprint/article, a linear solver is applied to both the predictions from the absolute model as well as the pairwise differences obtained from the differential model. Both are added as constraints and solved using least squares procedure.
• UI Improvements, major simplification in the frontend's layout, thereby simplifying the UX. It is now easier to get from compound to a solvent ranking.
• ported backend to use fastapi

v0.2 (snapshot used for preprint)

Corresponds to the state used in the preprint: https://chemrxiv.org/engage/chemrxiv/article-details/662f451f418a5379b0012795

Release: v0.1

pickled models trained on public data