Visual analysis of whole-slide images in histology is the basis for cancer diagnosis and the choice of an appropriate treatment. Treatments such as targeted therapies focus on a specific gene mutation, and automatically identifying such mutations solely based on images is still a challenge. Recently deep convolutional neural networks showed encouraging results on this problem (1). In this paper, we present HE2RNA (2), a new lightweight deep learning model to predict gene mutations in multiple cancers. This approach leverages pretrained networks (ResNet50) and image clustering using a simple averaging procedure. HE2RNA does not require any computationally expensive pre-selection step for the tumor regions contrary to the leading methods (1), (3). We evaluate our system on the public dataset The Cancer Genome Atlas (TCGA) and obtain comparable performances with the state-of-the-art Kather et al. (1), on Lung and Breast cancers. Moreover, we show promising performances on Low Grade Glioma cancer with an average AUC of 0.72 ([0.53 , 0.82]).