The presence of gut microbiome can have a significant impact on a person’s health and diseases. Gut microbiome that are collected from different locations tend to have different measurement even though they are the same samples due to the difference in equipment or handling creating batch effects. Rheumatoid arthritis is an autoimmune disease that affects multiple joints especially the finger leading to joint damage. A highly trained medical professional are often required to monitor the development of joint damage in a resource limited area which can reduce the efficiency to review the joint damage.In this MSc thesis, we show how using deep learning was able to classify longitudinal gut microbiome with missing data and ways to resolve missing data using imputation methods or padding methods. We also develop a deep learning for joint detection on rheumatoid arthritis patients, and that YOLOv5l6 is able to predict the bounding boxes of joints of rheumatoid arthritis patients with high performance even though YOLOv5l6 was trained on healthy joints. Pre-training with COCO dataset with YOLOv5l6 before training on the healthy joints was able to improve the performance of joint detection. Moreover, we also propose a deep learning autoencoder and extended LassoNet to classify disease status and remove batch effect in a single forward step through the analysis of the oral microbiome.