The inductive database vision (knowledge discovery as an extended querying process) has been suggested as a candidate formal framework for data mining. It has been studied as such within the EU funded cInQ IST-FET project (2001-2004) and it is now considered within the follow-up IQ IST-FET project (2005-2008). During this talk, we will survey the major results obtained within these projects so far. More precisely, we will consider our own practice of gene expression data analysis by means of data mining techniques (joint work with Celine Robardet, Jeremy Besson, Ruggero Pensa and Ieva Mitasiunaite). We will show that many interesting processes can be based on inductive queries which declaratively specify constraints on desired patterns (set patterns from 0/1 data encoding expression properties of genes in given samples, sequential patterns from ordered data like gene promoter sequences, global patterns like bi-partitions in gene expression data sets, etc). Supporting such processes needs for efficient solvers and we will illustrate how the targeted applications have given rise to nice scientific and technical problems, some of them solved, many others still open. The considered technical issues will concern our recent results w.r.t. constraint-based mining for bi-sets, fault-tolerant local pattern mining or constraint-based mining of bi-partitions.