We consider dynamic pricing and inventory control problems for multiple substitutable products in which the demand distribution and the customers' substitution behavior are not known a priori. Both stockout substitution and price-based substitution are considered, and unsatisfied demands are lost and unobservable. We develop data-driven algorithms that learn the demand and substitution information from past sales data and, concurrently, apply learned knowledge to maximize system profit. The algorithms are shown to be asymptotically optimal, and their convergence rates are also provided.