Non-fossil biomass gradually becomes a promising raw material in energy consumption structure for achieving carbon neutrality. Most of existing process modelling works mainly focus on fossil streams, challenges in molecular composition determination and digitalization restrain the application of existing works to biomass-derived materials. In this work, a molecular level molecular composition reconstruction framework is proposed, the framework covers representation of molecules and transformation between fraction information and mixture bulk properties. Molecular fingerprint method is introduced for structure description and a data-driven pure-component estimation method is integrated in the framework. Statistical method is implemented for parameter reduction in model optimisation. The accuracy and potential application of the methodology is evaluated by composition reconstruction of a diesel and a bio-oil sample, deviation between most of the typical measured and predicted properties are within 1 %. In addition, detailed molecular information is retained using the molecular fingerprint method, which makes it easier to integrate the proposed framework with existing frameworks.