An accurate diagnosis and profiling of tumour are critical to the best treatment choices for cancer patients. In addition to the cancer type and its aggressiveness, molecular heterogeneity also plays a vital role in treatment selection. MSI or MMR deficiency is one of the well-studied aberrations in terms of molecular changes. Colorectal cancer patients with MMR deficiency respond well to immunotherapy, hence assessment of the relevant molecular markers can assist clinicians in making optimal treatment selections for patients. Immunohistochemistry is one of the ways for identifying these molecular changes which requires additional sections of tumour tissue. Introduction of automated methods that can predict MSI or MMR status from a target image without the need for additional sections can substantially reduce the cost associated with it. In this work, we present our work on predicting MSI status in a two-stage process using a single target slide either stained with CK818 or H\&E. First, we train a multi-headed convolutional neural network model where each head is responsible for predicting one of the MMR protein expressions. To this end, we perform registration of MMR slides to the target slide as a pre-processing step. In the second stage, statistical features computed from the MMR prediction maps are used for the final MSI prediction. Our results demonstrate that MSI classification can be improved on incorporating fine-grained MMR labels in comparison to the previous approaches in which coarse labels (MSI/MSS) are utilised.