Micro-expressions are reflections of people's true feelings and motives, which attract an increasing number of researchers into the study of automatic facial micro-expression recognition. The short detection window, the subtle facial muscle movements, and the limited training samples make micro-expression recognition challenging. To this end, we propose a novel Identity-aware and Capsule-Enhanced Generative Adversarial Network with graph-based reasoning (ICE-GAN), introducing micro-expression synthesis as an auxiliary task to assist recognition. The generator produces synthetic faces with controllable micro-expressions and identity-aware features, whose long-ranged dependencies are captured through the graph reasoning module (GRM), and the discriminator detects the image authenticity and expression classes. Our ICE-GAN was evaluated on Micro-Expression Grand Challenge 2019 (MEGC2019) with a significant improvement (12.9%) over the winner and surpassed other state-of-the-art methods.