Hypothesis generation is an essential step in scientific discovery. It involves an analysis of existing evidence, the generation of a “theory” (or model) which leads to a hypothesis (about the unknown) tested by new experiments. The design of new drugs follows a similar scheme starting with an effective mining of a huge amount of collected experimental in vitro and in vivo data. These data often come from many different areas such as chemistry, biology, pharmacology, toxicology etc. and in various formats. Extracting the critical information is a challenging task that is performed by medicinal chemists and other scientists. Our goal is to assist the hypothesis generation and decision process by mimicking as much as possible the human reasoning. To do so, we have developed a decision-support system KEM (Knowledge Extraction and Management), based on the Galois lattices theory. KEM assists the scientist for efficiently generating and managing consistent hypothesis from experimental data. KEM® is a machine learning software that can guide multi-objective optimisation. We present here a few examples of data analysis by using KEM.