This paper explains our research and implementations of manual, automatic and deep annotations of provenance logs for e-Science insilico experiments. Compared to annotating general Web documents, annotations for scientific data require more sophisticated professional knowledge to recognize concepts from documents, and more complex text extraction and mapping mechanisms. A simple automatic annotation approach based on "lexicons" and a deep annotation implemented by semantically populating, translating and annotating provenance logs are introduced in this paper. We used COHSE (Conceptual Open Hypermedia Services Environment) to annotate and browse provenance logs from my Grid project, which are conceptually linked together as a hypertext Web of provenance logs and experiment resources, based on the associated conceptual metadata and reasoning over these metadata.