Depression, a widespread mental disorder, necessitates early and accurate diagnosis for effective intervention. Although clinical interviews are considered the gold standard, breakthroughs in artificial intelligence (AI) now provide automated depression prediction through the utilization of multimodal data, including text, audio, and video. This research utilizes the E-DAIC dataset, an augmented iteration of DAIC-WOZ, which includes 275 AI-facilitated clinical interviews that encompass facial expressions, auditory characteristics, and organized dialogues. We present a GMU-Transformer model that integrates Gated Multimodal Units (GMU) to improve the prediction of depression by attaining an MAE of 3.15 and an RMSE of 3.50.