Quantitative modelling of gene regulatory networks (GRNs) is still limited by data issues such as noise and the restricted length of available time series, creating an under-determination problem. However, large amounts of other types of biological data and knowledge are available, such as knockout experiments, annotations and so on, and it has been postulated that integration of these can improve model quality. However, integration has not been fully explored, to date. Here, we present a novel integrative framework for different types of data that aims to enhance model inference. This is based on evolutionary computation and uses different types of knowledge to introduce a novel customised initialisation and mutation operator and complex evaluation criteria, used to distinguish between candidate models. Specifically, the algorithm uses information from (i) knockout experiments, (ii) annotations of transcription factors, (iii) binding site motifs (expressed as position weight matrices) and (iv) DNA sequence of gene promoters, to drive the algorithm towards more plausible network structures. Further, the evaluation basis is also extended to include structure information included in these additional data. This framework is applied to both synthetic and real gene expression data. Models obtained by data integration display both quantitative and qualitative improvement.