The sequential data is today omnipresent and covers an important range of application domains. Sequence pattern mining allows to extract information and knowledge that can be of high added value. However, when the sequence data is rich in numerical data finer data mining methods are required to be able to extract more expressive knowledge representing the variability of numerical values and their possible interdependence. In this paper, we present a new method for the discovery of frequent gradual sequences represented by directed graphs (DAG), from a data sources of sequences in RDF (Resource Description Framework). The former ones, are first transformed into pog (partially ordered gradual sequences) DAGs labeled at the arcs and the nodes. Then, on these graphs, we apply an algorithm which mines and discovers the frequent pog subgraphs. Experiments on two real datasets in biology have shown the feasibility and the relevance of our approach.