This PHD thesis focuses on static analysis of programs by type inference in order to detect program errors before their execution. More precisely, we focus hear in the field of sub-typing, where program properties are described by sets of constraints of the form (t1 <= t2). Our verification mechanisms are based on the aggregation of sub-typing constraints and checking of their compatibility by saturation. The base language on which we define our type systems is an ML-like language provided with variants and pattern matching. We starts by defining a formalism to express our type systems thanks to inference rules. This formalism has the advantage to be sufficiently flexible to allow proving validity and termination properties of ...